to 0.2.13

Co-authored-by: yosagi <11404771+yosagi@users.noreply.github.com>"

.gitignore

@@ -54,21 +54,6 @@ coverage.xml
 # Translations
 *.mo
 *.pot
-# Django stuff:
-*.log
-local_settings.py
-db.sqlite3
-db.sqlite3-journal
-
-# Flask stuff:
-instance/
-.webassets-cache
-
-# Scrapy stuff:
-.scrapy
-
-# Sphinx documentation
-docs/_build/
 
 # PyBuilder
 target/
@@ -138,4 +123,5 @@ test_*.py
 launch.json
 .DS_Store
 test/*
-nllb-200-distilled-600M-ctranslate2/*
+nllb-200-distilled-600M-ctranslate2/*
+*.mp3

LICENSE

@@ -1,52 +1,210 @@
-# License
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
 
-## Main Software License
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
 
-MIT License
+   1. Definitions.
 
-Copyright (c) 2025 Quentin Fuxa.  
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
 
-Permission is hereby granted, free of charge, to any person obtaining a copy  
+      "Licensor" shall mean the copyright owner or entity authorized by
-of this software and associated documentation files (the "Software"), to deal  
+      the copyright owner that is granting the License.
-in the Software without restriction, including without limitation the rights  
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell  
-copies of the Software, and to permit persons to whom the Software is  
-furnished to do so, subject to the following conditions:
 
-The above copyright notice and this permission notice shall be included in all  
+      "Legal Entity" shall mean the union of the acting entity and all
-copies or substantial portions of the Software.
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
 
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR  
+      "You" (or "Your") shall mean an individual or Legal Entity
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,  
+      exercising permissions granted by this License.
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE  
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER  
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,  
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE  
-SOFTWARE.
 
-## SimulStreaming Backend License
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
 
-**When using the SimulStreaming backend (SimulWhisper), additional licensing terms apply:**
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
 
-SimulStreaming (https://github.com/ufal/SimulStreaming) is dual-licensed:
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
 
-### 🔹 Non-Commercial Use
+      "Derivative Works" shall mean any work, whether in Source or Object
-You may use SimulStreaming under the **PolyForm Noncommercial License 1.0.0** if you obtain the code through the GitHub repository. This license is **free of charge** and comes with **no obligations** for non-commercial users.
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
 
-### 🔸 Commercial Use
+      "Contribution" shall mean any work of authorship, including
-Understanding who uses SimulStreaming commercially helps improve and prioritize development. Therefore, **registration is required** for those who acquire a commercial license.
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
 
-Commercial licenses are planned to be **affordable** to SMEs and individuals. They are considering providing commercial licenses either for free or for a symbolic one-time fee, and may also provide additional support. You can share your preference via the [questionnaire](https://forms.cloud.microsoft.com/e/7tCxb4gJfB).
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
 
-You can also leave your contact [there](https://forms.cloud.microsoft.com/e/7tCxb4gJfB) to be notified when commercial licenses become available.
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
 
-**Contact for SimulStreaming licensing:**  
+   3. Grant of Patent License. Subject to the terms and conditions of
-[Dominik Macháček](https://ufal.mff.cuni.cz/dominik-machacek/), machacek@ufal.mff.cuni.cz
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
 
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2025 Quentin Fuxa
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
 ---
 
 ## Based on:
-- **whisper_streaming** by ÚFAL – MIT License – https://github.com/ufal/whisper_streaming. The original work by ÚFAL. License: https://github.com/ufal/whisper_streaming/blob/main/LICENSE  
+- **SimulWhisper** by Speech and Audio Technology LAB of Tsinghua University – Apache-2.0 – https://github.com/ufal/SimulStreaming
-- **silero-vad** by Snakers4 – MIT License – https://github.com/snakers4/silero-vad. The work by Snakers4 (silero-vad). License: https://github.com/snakers4/silero-vad/blob/f6b1294cb27590fb2452899df98fb234dfef1134/LICENSE  
+- **SimulStreaming** by ÚFAL – MIT License – https://github.com/ufal/SimulStreaming
-- **Diart** by juanmc2005 – MIT License – https://github.com/juanmc2005/diart. The work in Diart by juanmc2005. License: https://github.com/juanmc2005/diart/blob/main/LICENSE
+- **NeMo** by NVidia - Apache-2.0 - https://github.com/NVIDIA-NeMo/NeMo 
-- **SimulStreaming** by ÚFAL – Dual License (PolyForm Noncommercial License 1.0.0 / Commercial License) – https://github.com/ufal/SimulStreaming
+- **whisper_streaming** by ÚFAL – MIT License – https://github.com/ufal/whisper_streaming.
+- **silero-vad** by Snakers4 – MIT License – https://github.com/snakers4/silero-vad.
+- **Diart** by juanmc2005 – MIT License – https://github.com/juanmc2005/diart.

README.md

@@ -10,16 +10,16 @@
 <a href="https://pypi.org/project/whisperlivekit/"><img alt="PyPI Version" src="https://img.shields.io/pypi/v/whisperlivekit?color=g"></a>
 <a href="https://pepy.tech/project/whisperlivekit"><img alt="PyPI Downloads" src="https://static.pepy.tech/personalized-badge/whisperlivekit?period=total&units=international_system&left_color=grey&right_color=brightgreen&left_text=installations"></a>
 <a href="https://pypi.org/project/whisperlivekit/"><img alt="Python Versions" src="https://img.shields.io/badge/python-3.9--3.15-dark_green"></a>
-<a href="https://github.com/QuentinFuxa/WhisperLiveKit/blob/main/LICENSE"><img alt="License" src="https://img.shields.io/badge/License-MIT/Dual Licensed-dark_green"></a>
+<a href="https://github.com/QuentinFuxa/WhisperLiveKit/blob/main/LICENSE"><img alt="License" src="https://img.shields.io/badge/License-Apache 2.0-dark_green"></a>
 </p>
 
 
-Real-time speech transcription directly to your browser, with a ready-to-use backend+server and a simple frontend. ✨
+Real-time transcription directly to your browser, with a ready-to-use backend+server and a simple frontend.
 
 #### Powered by Leading Research:
 
-- [SimulStreaming](https://github.com/ufalSimulStreaming) (SOTA 2025) - Ultra-low latency transcription using [AlignAtt policy](https://arxiv.org/pdf/2305.11408)
+- Simul-[Whisper](https://github.com/backspacetg/simul_whisper)/[Streaming](https://github.com/ufal/SimulStreaming) (SOTA 2025) - Ultra-low latency transcription using [AlignAtt policy](https://arxiv.org/pdf/2305.11408)
-- [NLLB](https://arxiv.org/abs/2207.04672), ([distilled](https://huggingface.co/entai2965/nllb-200-distilled-600M-ctranslate2)) (2024) - Translation to more than 100 languages.
+- [NLLW](https://github.com/QuentinFuxa/NoLanguageLeftWaiting) (2025), based on [distilled](https://huggingface.co/entai2965/nllb-200-distilled-600M-ctranslate2) [NLLB](https://arxiv.org/abs/2207.04672) (2022, 2024) - Simulatenous translation from & to 200 languages.
 - [WhisperStreaming](https://github.com/ufal/whisper_streaming) (SOTA 2023) - Low latency transcription using [LocalAgreement policy](https://www.isca-archive.org/interspeech_2020/liu20s_interspeech.pdf)
 - [Streaming Sortformer](https://arxiv.org/abs/2507.18446) (SOTA 2025) - Advanced real-time speaker diarization
 - [Diart](https://github.com/juanmc2005/diart) (SOTA 2021) - Real-time speaker diarization
@@ -68,9 +68,9 @@ Go to `chrome-extension` for instructions.
 
 | Optional | `pip install` |
 |-----------|-------------|
-| **Speaker diarization with Sortformer** | `git+https://github.com/NVIDIA/NeMo.git@main#egg=nemo_toolkit[asr]` |
+| **Speaker diarization** | `git+https://github.com/NVIDIA/NeMo.git@main#egg=nemo_toolkit[asr]` |
-| **Apple Silicon optimized backend** | `mlx-whisper` |
+| **Apple Silicon optimizations** | `mlx-whisper` |
-| **NLLB Translation** | `huggingface_hub` & `transformers` |
+| **Translation** | `nllw` |
 | *[Not recommanded]*  Speaker diarization with Diart | `diart` |
 | *[Not recommanded]*  Original Whisper backend | `whisper` |
 | *[Not recommanded]*  Improved timestamps backend | `whisper-timestamped` |
@@ -139,10 +139,10 @@ async def websocket_endpoint(websocket: WebSocket):
 
 | Parameter | Description | Default |
 |-----------|-------------|---------|
-| `--model` | Whisper model size. List and recommandations [here](https://github.com/QuentinFuxa/WhisperLiveKit/blob/main/available_models.md) | `small` |
+| `--model` | Whisper model size. List and recommandations [here](https://github.com/QuentinFuxa/WhisperLiveKit/blob/main/docs/available_models.md) | `small` |
-| `--model-dir` | Directory containing Whisper model.bin and other files. Overrides `--model`. | `None` |
+| `--model-path` | .pt file/directory containing whisper model. Overrides `--model`. Recommandations [here](https://github.com/QuentinFuxa/WhisperLiveKit/blob/main/docs/models_compatible_formats.md) | `None` |
 | `--language` | List [here](https://github.com/QuentinFuxa/WhisperLiveKit/blob/main/whisperlivekit/simul_whisper/whisper/tokenizer.py). If you use `auto`, the model attempts to detect the language automatically, but it tends to bias towards English. | `auto` |
-| `--target-language` | If sets, activates translation using NLLB. Ex: `fr`. [118 languages available](https://github.com/QuentinFuxa/WhisperLiveKit/blob/main/whisperlivekit/translation/mapping_languages.py). If you want to translate to english, you should rather use `--task translate`, since Whisper can do it directly. | `None` |
+| `--target-language` | If sets, translate to using NLLB. Ex: `fr`. [200 languages available](https://github.com/QuentinFuxa/WhisperLiveKit/blob/main/docs/supported_languages.md). If you want to translate to english, you should rather use `--task translate`, since Whisper can do it directly. | `None` |
 | `--task` | Set to `translate` to translate *only* to english, using Whisper translation. | `transcribe` |
 | `--diarization` | Enable speaker identification | `False` |
 | `--backend` | Processing backend. You can switch to `faster-whisper` if  `simulstreaming` does not work correctly | `simulstreaming` |
@@ -182,7 +182,6 @@ async def websocket_endpoint(websocket: WebSocket):
 | `--init-prompt` | Initial prompt for the model | `None` |
 | `--static-init-prompt` | Static prompt that doesn't scroll | `None` |
 | `--max-context-tokens` | Maximum context tokens | `None` |
-| `--model-path` | Direct path to .pt model file. Download it if not found | `./base.pt` |
 | `--preload-model-count` | Optional. Number of models to preload in memory to speed up loading (set up to the expected number of concurrent users) | `1` |
 
 

docs/models_compatible_formats.md

@@ -0,0 +1,14 @@
+# Model Path Formats
+
+The `--model-path` parameter accepts:
+
+## File Path
+- **`.pt` format only** (required for AlignAtt policy decoder)
+
+## Directory Path (recommended)
+Must contain:
+- **`.pt` file** (required for decoder)
+
+May optionally contain:
+- **`.bin` file** - faster-whisper model for encoder (requires faster-whisper)
+- **`weights.npz`** or **`weights.safetensors`** - for encoder (requires whisper-mlx)

docs/supported_languages.md

@@ -0,0 +1,265 @@
+# Supported Languages
+
+WhisperLiveKit supports translation into **201 languages** from the FLORES-200 dataset through the NLLB (No Language Left Behind) translation system. 
+
+## How to Specify Languages
+
+You can specify languages in **three different ways**:
+
+1. **Language Name** (case-insensitive): `"English"`, `"French"`, `"Spanish"`
+2. **ISO Language Code**: `"en"`, `"fr"`, `"es"`
+3. **NLLB Code** (FLORES-200): `"eng_Latn"`, `"fra_Latn"`, `"spa_Latn"`
+
+## Usage Examples
+
+### Command Line
+```bash
+# Using language name
+whisperlivekit-server --target-language "French"
+
+# Using ISO code
+whisperlivekit-server --target-language fr
+
+# Using NLLB code
+whisperlivekit-server --target-language fra_Latn
+```
+
+### Python API
+```python
+from nllw.translation import get_language_info
+
+# Get language information by name
+lang_info = get_language_info("French")
+print(lang_info)
+# {'name': 'French', 'nllb': 'fra_Latn', 'language_code': 'fr'}
+
+# Get language information by ISO code
+lang_info = get_language_info("fr")
+
+# Get language information by NLLB code
+lang_info = get_language_info("fra_Latn")
+
+# All three return the same result
+```
+
+## Complete Language List
+
+The following table lists all 201 supported languages with their corresponding codes:
+
+| Language Name | ISO Code | NLLB Code |
+|---------------|----------|-----------|
+| Acehnese (Arabic script) | ace_Arab | ace_Arab |
+| Acehnese (Latin script) | ace_Latn | ace_Latn |
+| Mesopotamian Arabic | acm_Arab | acm_Arab |
+| Ta'izzi-Adeni Arabic | acq_Arab | acq_Arab |
+| Tunisian Arabic | aeb_Arab | aeb_Arab |
+| Afrikaans | af | afr_Latn |
+| South Levantine Arabic | ajp_Arab | ajp_Arab |
+| Akan | ak | aka_Latn |
+| Tosk Albanian | als | als_Latn |
+| Amharic | am | amh_Ethi |
+| North Levantine Arabic | apc_Arab | apc_Arab |
+| Modern Standard Arabic | ar | arb_Arab |
+| Modern Standard Arabic (Romanized) | arb_Latn | arb_Latn |
+| Najdi Arabic | ars_Arab | ars_Arab |
+| Moroccan Arabic | ary_Arab | ary_Arab |
+| Egyptian Arabic | arz_Arab | arz_Arab |
+| Assamese | as | asm_Beng |
+| Asturian | ast | ast_Latn |
+| Awadhi | awa | awa_Deva |
+| Central Aymara | ay | ayr_Latn |
+| South Azerbaijani | azb | azb_Arab |
+| North Azerbaijani | az | azj_Latn |
+| Bashkir | ba | bak_Cyrl |
+| Bambara | bm | bam_Latn |
+| Balinese | ban | ban_Latn |
+| Belarusian | be | bel_Cyrl |
+| Bemba | bem | bem_Latn |
+| Bengali | bn | ben_Beng |
+| Bhojpuri | bho | bho_Deva |
+| Banjar (Arabic script) | bjn_Arab | bjn_Arab |
+| Banjar (Latin script) | bjn_Latn | bjn_Latn |
+| Standard Tibetan | bo | bod_Tibt |
+| Bosnian | bs | bos_Latn |
+| Buginese | bug | bug_Latn |
+| Bulgarian | bg | bul_Cyrl |
+| Catalan | ca | cat_Latn |
+| Cebuano | ceb | ceb_Latn |
+| Czech | cs | ces_Latn |
+| Chokwe | cjk | cjk_Latn |
+| Central Kurdish | ckb | ckb_Arab |
+| Crimean Tatar | crh | crh_Latn |
+| Welsh | cy | cym_Latn |
+| Danish | da | dan_Latn |
+| German | de | deu_Latn |
+| Southwestern Dinka | dik | dik_Latn |
+| Dyula | dyu | dyu_Latn |
+| Dzongkha | dz | dzo_Tibt |
+| Greek | el | ell_Grek |
+| English | en | eng_Latn |
+| Esperanto | eo | epo_Latn |
+| Estonian | et | est_Latn |
+| Basque | eu | eus_Latn |
+| Ewe | ee | ewe_Latn |
+| Faroese | fo | fao_Latn |
+| Fijian | fj | fij_Latn |
+| Finnish | fi | fin_Latn |
+| Fon | fon | fon_Latn |
+| French | fr | fra_Latn |
+| Friulian | fur-IT | fur_Latn |
+| Nigerian Fulfulde | fuv | fuv_Latn |
+| West Central Oromo | om | gaz_Latn |
+| Scottish Gaelic | gd | gla_Latn |
+| Irish | ga-IE | gle_Latn |
+| Galician | gl | glg_Latn |
+| Guarani | gn | grn_Latn |
+| Gujarati | gu-IN | guj_Gujr |
+| Haitian Creole | ht | hat_Latn |
+| Hausa | ha | hau_Latn |
+| Hebrew | he | heb_Hebr |
+| Hindi | hi | hin_Deva |
+| Chhattisgarhi | hne | hne_Deva |
+| Croatian | hr | hrv_Latn |
+| Hungarian | hu | hun_Latn |
+| Armenian | hy-AM | hye_Armn |
+| Igbo | ig | ibo_Latn |
+| Ilocano | ilo | ilo_Latn |
+| Indonesian | id | ind_Latn |
+| Icelandic | is | isl_Latn |
+| Italian | it | ita_Latn |
+| Javanese | jv | jav_Latn |
+| Japanese | ja | jpn_Jpan |
+| Kabyle | kab | kab_Latn |
+| Jingpho | kac | kac_Latn |
+| Kamba | kam | kam_Latn |
+| Kannada | kn | kan_Knda |
+| Kashmiri (Arabic script) | kas_Arab | kas_Arab |
+| Kashmiri (Devanagari script) | kas_Deva | kas_Deva |
+| Georgian | ka | kat_Geor |
+| Kazakh | kk | kaz_Cyrl |
+| Kabiyè | kbp | kbp_Latn |
+| Kabuverdianu | kea | kea_Latn |
+| Halh Mongolian | mn | khk_Cyrl |
+| Khmer | km | khm_Khmr |
+| Kikuyu | ki | kik_Latn |
+| Kinyarwanda | rw | kin_Latn |
+| Kyrgyz | ky | kir_Cyrl |
+| Kimbundu | kmb | kmb_Latn |
+| Northern Kurdish | kmr | kmr_Latn |
+| Central Kanuri (Arabic script) | knc_Arab | knc_Arab |
+| Central Kanuri (Latin script) | knc_Latn | knc_Latn |
+| Kikongo | kg | kon_Latn |
+| Korean | ko | kor_Hang |
+| Lao | lo | lao_Laoo |
+| Ligurian | lij | lij_Latn |
+| Limburgish | li | lim_Latn |
+| Lingala | ln | lin_Latn |
+| Lithuanian | lt | lit_Latn |
+| Lombard | lmo | lmo_Latn |
+| Latgalian | ltg | ltg_Latn |
+| Luxembourgish | lb | ltz_Latn |
+| Luba-Kasai | lua | lua_Latn |
+| Ganda | lg | lug_Latn |
+| Luo | luo | luo_Latn |
+| Mizo | lus | lus_Latn |
+| Standard Latvian | lv | lvs_Latn |
+| Magahi | mag | mag_Deva |
+| Maithili | mai | mai_Deva |
+| Malayalam | ml-IN | mal_Mlym |
+| Marathi | mr | mar_Deva |
+| Minangkabau (Arabic script) | min_Arab | min_Arab |
+| Minangkabau (Latin script) | min_Latn | min_Latn |
+| Macedonian | mk | mkd_Cyrl |
+| Maltese | mt | mlt_Latn |
+| Meitei (Bengali script) | mni | mni_Beng |
+| Mossi | mos | mos_Latn |
+| Maori | mi | mri_Latn |
+| Burmese | my | mya_Mymr |
+| Dutch | nl | nld_Latn |
+| Norwegian Nynorsk | nn-NO | nno_Latn |
+| Norwegian Bokmål | nb | nob_Latn |
+| Nepali | ne-NP | npi_Deva |
+| Northern Sotho | nso | nso_Latn |
+| Nuer | nus | nus_Latn |
+| Nyanja | ny | nya_Latn |
+| Occitan | oc | oci_Latn |
+| Odia | or | ory_Orya |
+| Pangasinan | pag | pag_Latn |
+| Eastern Panjabi | pa | pan_Guru |
+| Papiamento | pap | pap_Latn |
+| Southern Pashto | pbt | pbt_Arab |
+| Western Persian | fa | pes_Arab |
+| Plateau Malagasy | mg | plt_Latn |
+| Polish | pl | pol_Latn |
+| Portuguese | pt-PT | por_Latn |
+| Dari | fa-AF | prs_Arab |
+| Ayacucho Quechua | qu | quy_Latn |
+| Romanian | ro | ron_Latn |
+| Rundi | rn | run_Latn |
+| Russian | ru | rus_Cyrl |
+| Sango | sg | sag_Latn |
+| Sanskrit | sa | san_Deva |
+| Santali | sat | sat_Olck |
+| Sicilian | scn | scn_Latn |
+| Shan | shn | shn_Mymr |
+| Sinhala | si-LK | sin_Sinh |
+| Slovak | sk | slk_Latn |
+| Slovenian | sl | slv_Latn |
+| Samoan | sm | smo_Latn |
+| Shona | sn | sna_Latn |
+| Sindhi | sd | snd_Arab |
+| Somali | so | som_Latn |
+| Southern Sotho | st | sot_Latn |
+| Spanish | es-ES | spa_Latn |
+| Sardinian | sc | srd_Latn |
+| Serbian | sr | srp_Cyrl |
+| Swati | ss | ssw_Latn |
+| Sundanese | su | sun_Latn |
+| Swedish | sv-SE | swe_Latn |
+| Swahili | sw | swh_Latn |
+| Silesian | szl | szl_Latn |
+| Tamil | ta | tam_Taml |
+| Tamasheq (Latin script) | taq_Latn | taq_Latn |
+| Tamasheq (Tifinagh script) | taq_Tfng | taq_Tfng |
+| Tatar | tt-RU | tat_Cyrl |
+| Telugu | te | tel_Telu |
+| Tajik | tg | tgk_Cyrl |
+| Tagalog | tl | tgl_Latn |
+| Thai | th | tha_Thai |
+| Tigrinya | ti | tir_Ethi |
+| Tok Pisin | tpi | tpi_Latn |
+| Tswana | tn | tsn_Latn |
+| Tsonga | ts | tso_Latn |
+| Turkmen | tk | tuk_Latn |
+| Tumbuka | tum | tum_Latn |
+| Turkish | tr | tur_Latn |
+| Twi | tw | twi_Latn |
+| Central Atlas Tamazight | tzm | tzm_Tfng |
+| Uyghur | ug | uig_Arab |
+| Ukrainian | uk | ukr_Cyrl |
+| Umbundu | umb | umb_Latn |
+| Urdu | ur | urd_Arab |
+| Northern Uzbek | uz | uzn_Latn |
+| Venetian | vec | vec_Latn |
+| Vietnamese | vi | vie_Latn |
+| Waray | war | war_Latn |
+| Wolof | wo | wol_Latn |
+| Xhosa | xh | xho_Latn |
+| Eastern Yiddish | yi | ydd_Hebr |
+| Yoruba | yo | yor_Latn |
+| Yue Chinese | yue | yue_Hant |
+| Chinese (Simplified) | zh-CN | zho_Hans |
+| Chinese (Traditional) | zh-TW | zho_Hant |
+| Standard Malay | ms | zsm_Latn |
+| Zulu | zu | zul_Latn |
+
+## Special Features
+
+### Multiple Script Support
+Several languages are available in multiple scripts (e.g., Arabic and Latin):
+- **Acehnese**: Arabic (`ace_Arab`) and Latin (`ace_Latn`)
+- **Banjar**: Arabic (`bjn_Arab`) and Latin (`bjn_Latn`)
+- **Kashmiri**: Arabic (`kas_Arab`) and Devanagari (`kas_Deva`)
+- **Minangkabau**: Arabic (`min_Arab`) and Latin (`min_Latn`)
+- **Tamasheq**: Latin (`taq_Latn`) and Tifinagh (`taq_Tfng`)
+- **Central Kanuri**: Arabic (`knc_Arab`) and Latin (`knc_Latn`)

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "whisperlivekit"
-version = "0.2.12"
+version = "0.2.13"
 description = "Real-time speech-to-text with speaker diarization using Whisper"
 readme = "README.md"
 authors = [
@@ -41,7 +41,8 @@ dependencies = [
 ]
 
 [project.optional-dependencies]
-sentence = ["mosestokenizer", "wtpsplit"]
+translation = ["nllw"]
+sentence_tokenizer = ["mosestokenizer", "wtpsplit"]
 
 [project.urls]
 Homepage = "https://github.com/QuentinFuxa/WhisperLiveKit"
@@ -50,8 +51,19 @@ Homepage = "https://github.com/QuentinFuxa/WhisperLiveKit"
 whisperlivekit-server = "whisperlivekit.basic_server:main"
 
 [tool.setuptools]
-packages = ["whisperlivekit", "whisperlivekit.diarization", "whisperlivekit.simul_whisper", "whisperlivekit.simul_whisper.whisper", "whisperlivekit.simul_whisper.whisper.assets", "whisperlivekit.simul_whisper.whisper.normalizers", "whisperlivekit.web", "whisperlivekit.whisper_streaming_custom", "whisperlivekit.translation"]
+packages = [
+    "whisperlivekit",
+    "whisperlivekit.diarization",
+    "whisperlivekit.simul_whisper",
+    "whisperlivekit.simul_whisper.whisper",
+    "whisperlivekit.simul_whisper.whisper.assets",
+    "whisperlivekit.simul_whisper.whisper.normalizers",
+    "whisperlivekit.web",
+    "whisperlivekit.whisper_streaming_custom",
+    "whisperlivekit.vad_models"
+]
 
 [tool.setuptools.package-data]
 whisperlivekit = ["web/*.html", "web/*.css", "web/*.js", "web/src/*.svg"]
 "whisperlivekit.simul_whisper.whisper.assets" = ["*.tiktoken", "*.npz"]
+"whisperlivekit.vad_models" = ["*.jit", "*.onnx"]

whisperlivekit/audio_processor.py

@@ -67,20 +67,17 @@ class AudioProcessor:
         self.is_stopping = False
         self.silence = False
         self.silence_duration = 0.0
-        self.tokens = []
+        self.state = State()
-        self.last_validated_token = 0
-        self.translated_segments = []
-        self.buffer_transcription = Transcript()
-        self.end_buffer = 0
-        self.end_attributed_speaker = 0
         self.lock = asyncio.Lock()
-        self.beg_loop = 0.0 #to deal with a potential little lag at the websocket initialization, this is now set in process_audio
         self.sep = " "  # Default separator
         self.last_response_content = FrontData()
         self.last_detected_speaker = None
         self.speaker_languages = {}
         self.diarization_before_transcription = False
 
+        self.segments = []
+        
+
         if self.diarization_before_transcription:
             self.cumulative_pcm = []
             self.last_start = 0.0
@@ -138,8 +135,8 @@ class AudioProcessor:
     async def add_dummy_token(self):
         """Placeholder token when no transcription is available."""
         async with self.lock:
-            current_time = time() - self.beg_loop
+            current_time = time() - self.state.beg_loop
-            self.tokens.append(ASRToken(
+            self.state.tokens.append(ASRToken(
                 start=current_time, end=current_time + 1,
                 text=".", speaker=-1, is_dummy=True
             ))
@@ -149,35 +146,19 @@ class AudioProcessor:
         async with self.lock:
             current_time = time()
             
-            # Calculate remaining times
             remaining_transcription = 0
-            if self.end_buffer > 0:
+            if self.state.end_buffer > 0:
-                remaining_transcription = max(0, round(current_time - self.beg_loop - self.end_buffer, 1))
+                remaining_transcription = max(0, round(current_time - self.state.beg_loop - self.state.end_buffer, 1))
                 
             remaining_diarization = 0
-            if self.tokens:
+            if self.state.tokens:
-                latest_end = max(self.end_buffer, self.tokens[-1].end if self.tokens else 0)
+                latest_end = max(self.state.end_buffer, self.state.tokens[-1].end if self.state.tokens else 0)
-                remaining_diarization = max(0, round(latest_end - self.end_attributed_speaker, 1))
+                remaining_diarization = max(0, round(latest_end - self.state.end_attributed_speaker, 1))
                 
-            return State(
+            self.state.remaining_time_transcription = remaining_transcription
-                tokens=self.tokens.copy(),
+            self.state.remaining_time_diarization = remaining_diarization
-                last_validated_token=self.last_validated_token,
-                translated_segments=self.translated_segments.copy(),
-                buffer_transcription=self.buffer_transcription,
-                end_buffer=self.end_buffer,
-                end_attributed_speaker=self.end_attributed_speaker,
-                remaining_time_transcription=remaining_transcription,
-                remaining_time_diarization=remaining_diarization
-            )
             
-    async def reset(self):
+            return self.state
-        """Reset all state variables to initial values."""
-        async with self.lock:
-            self.tokens = []
-            self.translated_segments = []
-            self.buffer_transcription = Transcript()
-            self.end_buffer = self.end_attributed_speaker = 0
-            self.beg_loop = time()
 
     async def ffmpeg_stdout_reader(self):
         """Read audio data from FFmpeg stdout and process it into the PCM pipeline."""
@@ -242,15 +223,15 @@ class AudioProcessor:
                     break
 
                 asr_internal_buffer_duration_s = len(getattr(self.transcription, 'audio_buffer', [])) / self.transcription.SAMPLING_RATE
-                transcription_lag_s = max(0.0, time() - self.beg_loop - self.end_buffer)
+                transcription_lag_s = max(0.0, time() - self.state.beg_loop - self.state.end_buffer)
                 asr_processing_logs = f"internal_buffer={asr_internal_buffer_duration_s:.2f}s | lag={transcription_lag_s:.2f}s |"
                 if type(item) is Silence:
                     asr_processing_logs += f" + Silence of = {item.duration:.2f}s"
-                    if self.tokens:
+                    if self.state.tokens:
-                        asr_processing_logs += f" | last_end = {self.tokens[-1].end} |"
+                        asr_processing_logs += f" | last_end = {self.state.tokens[-1].end} |"
                     logger.info(asr_processing_logs)
                     cumulative_pcm_duration_stream_time += item.duration
-                    self.transcription.insert_silence(item.duration, self.tokens[-1].end if self.tokens else 0)
+                    self.transcription.insert_silence(item.duration, self.state.tokens[-1].end if self.state.tokens else 0)
                     continue
                 elif isinstance(item, ChangeSpeaker):
                     self.transcription.new_speaker(item)
@@ -274,7 +255,7 @@ class AudioProcessor:
                     if buffer_text.startswith(validated_text):
                         _buffer_transcript.text = buffer_text[len(validated_text):].lstrip()
 
-                candidate_end_times = [self.end_buffer]
+                candidate_end_times = [self.state.end_buffer]
 
                 if new_tokens:
                     candidate_end_times.append(new_tokens[-1].end)
@@ -285,9 +266,9 @@ class AudioProcessor:
                 candidate_end_times.append(current_audio_processed_upto)
                 
                 async with self.lock:
-                    self.tokens.extend(new_tokens)
+                    self.state.tokens.extend(new_tokens)
-                    self.buffer_transcription = _buffer_transcript
+                    self.state.buffer_transcription = _buffer_transcript
-                    self.end_buffer = max(candidate_end_times)
+                    self.state.end_buffer = max(candidate_end_times)
                 
                 if self.translation_queue:
                     for token in new_tokens:
@@ -360,12 +341,12 @@ class AudioProcessor:
                             self.last_end = last_segment.end
                 elif not self.diarization_before_transcription:           
                     async with self.lock:
-                        self.tokens = diarization_obj.assign_speakers_to_tokens(
+                        self.state.tokens = diarization_obj.assign_speakers_to_tokens(
-                            self.tokens,
+                            self.state.tokens,
                             use_punctuation_split=self.args.punctuation_split
                         )
-                if len(self.tokens) > 0:
+                if len(self.state.tokens) > 0:
-                    self.end_attributed_speaker = max(self.tokens[-1].end, self.end_attributed_speaker)
+                    self.state.end_attributed_speaker = max(self.state.tokens[-1].end, self.state.end_attributed_speaker)
                 self.diarization_queue.task_done()
                 
             except Exception as e:
@@ -406,7 +387,10 @@ class AudioProcessor:
                         tokens_to_process.append(additional_token)                
                 if tokens_to_process:
                     self.translation.insert_tokens(tokens_to_process)
-                    self.translated_segments = await asyncio.to_thread(self.translation.process)
+                    translation_validated_segments, translation_buffer = await asyncio.to_thread(self.translation.process)
+                    async with self.lock:
+                        self.state.translation_validated_segments = translation_validated_segments
+                        self.state.translation_buffer = translation_buffer
                 self.translation_queue.task_done()
                 for _ in additional_tokens:
                     self.translation_queue.task_done()
@@ -437,11 +421,9 @@ class AudioProcessor:
 
                 state = await self.get_current_state()
                 
-                
                 lines, undiarized_text = format_output(
                     state,
                     self.silence,
-                    current_time = time() - self.beg_loop,
                     args = self.args,
                     sep=self.sep
                 )
@@ -455,7 +437,7 @@ class AudioProcessor:
                     buffer_diarization = self.sep.join(undiarized_text)
 
                     async with self.lock:
-                        self.end_attributed_speaker = state.end_attributed_speaker
+                        self.state.end_attributed_speaker = state.end_attributed_speaker
                 
                 response_status = "active_transcription"
                 if not state.tokens and not buffer_transcription and not buffer_diarization:
@@ -482,23 +464,14 @@ class AudioProcessor:
                     yield response
                     self.last_response_content = response
                 
-                # Check for termination condition
+                if self.is_stopping and self.transcription_task and self.transcription_task.done() and self.diarization_task and self.diarization_task.done():
-                if self.is_stopping:
+                    logger.info("Results formatter: All upstream processors are done and in stopping state. Terminating.")
-                    all_processors_done = True
+                    return
-                    if self.args.transcription and self.transcription_task and not self.transcription_task.done():
-                        all_processors_done = False
-                    if self.args.diarization and self.diarization_task and not self.diarization_task.done():
-                        all_processors_done = False
-                    
-                    if all_processors_done:
-                        logger.info("Results formatter: All upstream processors are done and in stopping state. Terminating.")
-                        return
                 
                 await asyncio.sleep(0.05)
                 
             except Exception as e:
-                logger.warning(f"Exception in results_formatter: {e}")
+                logger.warning(f"Exception in results_formatter. Traceback: {traceback.format_exc()}")
-                logger.warning(f"Traceback: {traceback.format_exc()}")
                 await asyncio.sleep(0.5)
         
     async def create_tasks(self):
@@ -590,8 +563,8 @@ class AudioProcessor:
     async def process_audio(self, message):
         """Process incoming audio data."""
 
-        if not self.beg_loop:
+        if not self.state.beg_loop:
-            self.beg_loop = time()
+            self.state.beg_loop = time()
 
         if not message:
             logger.info("Empty audio message received, initiating stop sequence.")

whisperlivekit/core.py

@@ -33,6 +33,7 @@ class TranscriptionEngine:
             "punctuation_split": False,
             "target_language": "",
             "vac": True,
+            "vac_onnx": False,
             "vac_chunk_size": 0.04,
             "log_level": "DEBUG",
             "ssl_certfile": None,
@@ -75,8 +76,10 @@ class TranscriptionEngine:
         self.vac_model = None
         
         if self.args.vac:
-            import torch
+            from whisperlivekit.silero_vad_iterator import load_silero_vad
-            self.vac_model, _ = torch.hub.load(repo_or_dir="snakers4/silero-vad", model="silero_vad")            
+            # Use ONNX if specified, otherwise use JIT (default)
+            use_onnx = kwargs.get('vac_onnx', False)
+            self.vac_model = load_silero_vad(onnx=use_onnx)
         
         if self.args.transcription:
             if self.args.backend == "simulstreaming": 
@@ -138,9 +141,12 @@ class TranscriptionEngine:
             if self.args.lan == 'auto' and self.args.backend != "simulstreaming":
                 raise Exception('Translation cannot be set with language auto when transcription backend is not simulstreaming')
             else:
-                from whisperlivekit.translation.translation import load_model
+                try:
+                    from nllw import load_model
+                except:
+                    raise Exception('To use translation, you must install nllw: `pip install nllw`')
                 translation_params = { 
-                    "nllb_backend": "ctranslate2",
+                    "nllb_backend": "transformers",
                     "nllb_size": "600M"
                 }
                 translation_params = update_with_kwargs(translation_params, kwargs)
@@ -172,5 +178,5 @@ def online_translation_factory(args, translation_model):
     #should be at speaker level in the future:
     #one shared nllb model for all speaker
     #one tokenizer per speaker/language
-    from whisperlivekit.translation.translation import OnlineTranslation
+    from nllw import OnlineTranslation
-    return OnlineTranslation(translation_model, [args.lan], [args.target_language])
+    return OnlineTranslation(translation_model, [args.lan], [args.target_language])

whisperlivekit/parse_args.py

@@ -300,7 +300,7 @@ def parse_args():
     simulstreaming_group.add_argument(
         "--nllb-backend",
         type=str,
-        default="ctranslate2",
+        default="transformers",
         help="transformers or ctranslate2",
     )
     

whisperlivekit/remove_silences.py

@@ -1,4 +1,5 @@
 from whisperlivekit.timed_objects import ASRToken
+from time import time
 import re
 
 MIN_SILENCE_DURATION = 4 #in seconds
@@ -77,7 +78,8 @@ def no_token_to_silence(tokens):
             new_tokens.append(token)
     return new_tokens
             
-def ends_with_silence(tokens, current_time, vac_detected_silence):
+def ends_with_silence(tokens, beg_loop, vac_detected_silence):
+    current_time = time() - (beg_loop if beg_loop else 0.0)
     last_token = tokens[-1]
     if  vac_detected_silence or (current_time - last_token.end >= END_SILENCE_DURATION):
         if last_token.speaker == -2:
@@ -94,11 +96,11 @@ def ends_with_silence(tokens, current_time, vac_detected_silence):
     return tokens
     
 
-def handle_silences(tokens, current_time, vac_detected_silence):
+def handle_silences(tokens, beg_loop, vac_detected_silence):
     if not tokens:
         return []
     tokens = blank_to_silence(tokens) #useful for simulstreaming backend which tends to generate [BLANK_AUDIO] text
     tokens = no_token_to_silence(tokens)
-    tokens = ends_with_silence(tokens, current_time, vac_detected_silence)
+    tokens = ends_with_silence(tokens, beg_loop, vac_detected_silence)
     return tokens
      

whisperlivekit/results_formater.py

@@ -52,16 +52,17 @@ def append_token_to_last_line(lines, sep, token):
             lines[-1].detected_language = token.detected_language
             
 
-def format_output(state, silence, current_time, args, sep):
+def format_output(state, silence, args, sep):
     diarization = args.diarization
     disable_punctuation_split = args.disable_punctuation_split
     tokens = state.tokens
-    translated_segments = state.translated_segments # Here we will attribute the speakers only based on the timestamps of the segments
+    translation_validated_segments = state.translation_validated_segments # Here we will attribute the speakers only based on the timestamps of the segments
+    translation_buffer = state.translation_buffer
     last_validated_token = state.last_validated_token
     
     previous_speaker = 1
     undiarized_text = []
-    tokens = handle_silences(tokens, current_time, silence)
+    tokens = handle_silences(tokens, state.beg_loop, silence)
     last_punctuation = None
     for i, token in enumerate(tokens[last_validated_token:]):
         speaker = int(token.speaker)
@@ -71,13 +72,6 @@ def format_output(state, silence, current_time, args, sep):
                 token.corrected_speaker = 1
                 token.validated_speaker = True
         else:
-            # if token.end > end_attributed_speaker and token.speaker != -2:
-            #     if tokens[-1].speaker == -2:  #if it finishes by a silence, we want to append the undiarized text to the last speaker.
-            #         token.corrected_speaker = previous_speaker
-            #     else:
-            #         undiarized_text.append(token.text)
-            #         continue
-            # else:
                 if is_punctuation(token):
                     last_punctuation = i  
                 
@@ -123,9 +117,9 @@ def format_output(state, silence, current_time, args, sep):
 
         previous_speaker = token.corrected_speaker        
 
-    if lines and translated_segments:
+    if lines:
         unassigned_translated_segments = []
-        for ts in translated_segments:
+        for ts in translation_validated_segments:
             assigned = False
             for line in lines:
                 if ts and ts.overlaps_with(line):

whisperlivekit/silero_vad_iterator.py

@@ -1,27 +1,182 @@
 import torch
+import numpy as np
+import warnings
+from pathlib import Path
 
-# This is copied from silero-vad's vad_utils.py:
+"""
-# https://github.com/snakers4/silero-vad/blob/f6b1294cb27590fb2452899df98fb234dfef1134/utils_vad.py#L340
+Code is adapted from silero-vad v6: https://github.com/snakers4/silero-vad
-# (except changed defaults)
+"""
 
-# Their licence is MIT, same as ours: https://github.com/snakers4/silero-vad/blob/f6b1294cb27590fb2452899df98fb234dfef1134/LICENSE
+def init_jit_model(model_path: str, device=torch.device('cpu')):
+    """Load a JIT model from file."""
+    model = torch.jit.load(model_path, map_location=device)
+    model.eval()
+    return model
+
+
+class OnnxWrapper():
+    """ONNX Runtime wrapper for Silero VAD model."""
+
+    def __init__(self, path, force_onnx_cpu=False):
+        global np
+        import numpy as np
+        import onnxruntime
+
+        opts = onnxruntime.SessionOptions()
+        opts.inter_op_num_threads = 1
+        opts.intra_op_num_threads = 1
+
+        if force_onnx_cpu and 'CPUExecutionProvider' in onnxruntime.get_available_providers():
+            self.session = onnxruntime.InferenceSession(path, providers=['CPUExecutionProvider'], sess_options=opts)
+        else:
+            self.session = onnxruntime.InferenceSession(path, sess_options=opts)
+
+        self.reset_states()
+        if '16k' in path:
+            warnings.warn('This model support only 16000 sampling rate!')
+            self.sample_rates = [16000]
+        else:
+            self.sample_rates = [8000, 16000]
+
+    def _validate_input(self, x, sr: int):
+        if x.dim() == 1:
+            x = x.unsqueeze(0)
+        if x.dim() > 2:
+            raise ValueError(f"Too many dimensions for input audio chunk {x.dim()}")
+
+        if sr != 16000 and (sr % 16000 == 0):
+            step = sr // 16000
+            x = x[:,::step]
+            sr = 16000
+
+        if sr not in self.sample_rates:
+            raise ValueError(f"Supported sampling rates: {self.sample_rates} (or multiply of 16000)")
+        if sr / x.shape[1] > 31.25:
+            raise ValueError("Input audio chunk is too short")
+
+        return x, sr
+
+    def reset_states(self, batch_size=1):
+        self._state = torch.zeros((2, batch_size, 128)).float()
+        self._context = torch.zeros(0)
+        self._last_sr = 0
+        self._last_batch_size = 0
+
+    def __call__(self, x, sr: int):
+
+        x, sr = self._validate_input(x, sr)
+        num_samples = 512 if sr == 16000 else 256
+
+        if x.shape[-1] != num_samples:
+            raise ValueError(f"Provided number of samples is {x.shape[-1]} (Supported values: 256 for 8000 sample rate, 512 for 16000)")
+
+        batch_size = x.shape[0]
+        context_size = 64 if sr == 16000 else 32
+
+        if not self._last_batch_size:
+            self.reset_states(batch_size)
+        if (self._last_sr) and (self._last_sr != sr):
+            self.reset_states(batch_size)
+        if (self._last_batch_size) and (self._last_batch_size != batch_size):
+            self.reset_states(batch_size)
+
+        if not len(self._context):
+            self._context = torch.zeros(batch_size, context_size)
+
+        x = torch.cat([self._context, x], dim=1)
+        if sr in [8000, 16000]:
+            ort_inputs = {'input': x.numpy(), 'state': self._state.numpy(), 'sr': np.array(sr, dtype='int64')}
+            ort_outs = self.session.run(None, ort_inputs)
+            out, state = ort_outs
+            self._state = torch.from_numpy(state)
+        else:
+            raise ValueError()
+
+        self._context = x[..., -context_size:]
+        self._last_sr = sr
+        self._last_batch_size = batch_size
+
+        out = torch.from_numpy(out)
+        return out
+
+
+def load_silero_vad(model_path: str = None, onnx: bool = False, opset_version: int = 16):
+    """
+    Load Silero VAD model (JIT or ONNX).
+    
+    Parameters
+    ----------
+    model_path : str, optional
+        Path to model file. If None, uses default bundled model.
+    onnx : bool, default False
+        Whether to use ONNX runtime (requires onnxruntime package).
+    opset_version : int, default 16
+        ONNX opset version (15 or 16). Only used if onnx=True.
+    
+    Returns
+    -------
+    model
+        Loaded VAD model (JIT or ONNX wrapper)
+    """
+    available_ops = [15, 16]
+    if onnx and opset_version not in available_ops:
+        raise Exception(f'Available ONNX opset_version: {available_ops}')
+    if model_path is None:
+        current_dir = Path(__file__).parent
+        data_dir = current_dir / 'vad_models'
+        
+        if onnx:
+            if opset_version == 16:
+                model_name = 'silero_vad.onnx'
+            else:
+                model_name = f'silero_vad_16k_op{opset_version}.onnx'
+        else:
+            model_name = 'silero_vad.jit'
+        
+        model_path = data_dir / model_name
+        
+        if not model_path.exists():
+            raise FileNotFoundError(
+                f"Model file not found: {model_path}\n"
+                f"Please ensure the whisperlivekit/vad_models/ directory contains the model files."
+            )
+    else:
+        model_path = Path(model_path)
+    if onnx:
+        try:
+            model = OnnxWrapper(str(model_path), force_onnx_cpu=True)
+        except ImportError:
+            raise ImportError(
+                "ONNX runtime not available. Install with: pip install onnxruntime\n"
+                "Or use JIT model by setting onnx=False"
+            )
+    else:
+        model = init_jit_model(str(model_path))
+
+    return model
 
 
 class VADIterator:
-    def __init__(
+    """
-        self,
+    Voice Activity Detection iterator for streaming audio.
-        model,
+    
-        threshold: float = 0.5,
+    This is the Silero VAD v6 implementation.
-        sampling_rate: int = 16000,
+    """
-        min_silence_duration_ms: int = 500,  # makes sense on one recording that I checked
+    
-        speech_pad_ms: int = 100,  # same
+    def __init__(self,
-    ):
+                 model,
+                 threshold: float = 0.5,
+                 sampling_rate: int = 16000,
+                 min_silence_duration_ms: int = 100,
+                 speech_pad_ms: int = 30
+                 ):
+
         """
         Class for stream imitation
 
         Parameters
         ----------
-        model: preloaded .jit silero VAD model
+        model: preloaded .jit/.onnx silero VAD model
 
         threshold: float (default - 0.5)
             Speech threshold. Silero VAD outputs speech probabilities for each audio chunk, probabilities ABOVE this value are considered as SPEECH.
@@ -42,9 +197,7 @@ class VADIterator:
         self.sampling_rate = sampling_rate
 
         if sampling_rate not in [8000, 16000]:
-            raise ValueError(
+            raise ValueError('VADIterator does not support sampling rates other than [8000, 16000]')
-                "VADIterator does not support sampling rates other than [8000, 16000]"
-            )
 
         self.min_silence_samples = sampling_rate * min_silence_duration_ms / 1000
         self.speech_pad_samples = sampling_rate * speech_pad_ms / 1000
@@ -57,13 +210,17 @@ class VADIterator:
         self.temp_end = 0
         self.current_sample = 0
 
-    def __call__(self, x, return_seconds=False):
+    @torch.no_grad()
+    def __call__(self, x, return_seconds=False, time_resolution: int = 1):
         """
         x: torch.Tensor
             audio chunk (see examples in repo)
 
         return_seconds: bool (default - False)
             whether return timestamps in seconds (default - samples)
+
+        time_resolution: int (default - 1)
+            time resolution of speech coordinates when requested as seconds
         """
 
         if not torch.is_tensor(x):
@@ -82,14 +239,8 @@ class VADIterator:
 
         if (speech_prob >= self.threshold) and not self.triggered:
             self.triggered = True
-            speech_start = self.current_sample - self.speech_pad_samples
+            speech_start = max(0, self.current_sample - self.speech_pad_samples - window_size_samples)
-            return {
+            return {'start': int(speech_start) if not return_seconds else round(speech_start / self.sampling_rate, time_resolution)}
-                "start": (
-                    int(speech_start)
-                    if not return_seconds
-                    else round(speech_start / self.sampling_rate, 1)
-                )
-            }
 
         if (speech_prob < self.threshold - 0.15) and self.triggered:
             if not self.temp_end:
@@ -97,30 +248,17 @@ class VADIterator:
             if self.current_sample - self.temp_end < self.min_silence_samples:
                 return None
             else:
-                speech_end = self.temp_end + self.speech_pad_samples
+                speech_end = self.temp_end + self.speech_pad_samples - window_size_samples
                 self.temp_end = 0
                 self.triggered = False
-                return {
+                return {'end': int(speech_end) if not return_seconds else round(speech_end / self.sampling_rate, time_resolution)}
-                    "end": (
-                        int(speech_end)
-                        if not return_seconds
-                        else round(speech_end / self.sampling_rate, 1)
-                    )
-                }
 
         return None
 
 
-#######################
-# because Silero now requires exactly 512-sized audio chunks
-
-import numpy as np
-
-
 class FixedVADIterator(VADIterator):
-    """It fixes VADIterator by allowing to process any audio length, not only exactly 512 frames at once.
+    """
-    If audio to be processed at once is long and multiple voiced segments detected,
+    Fixed VAD Iterator that handles variable-length audio chunks, not only exactly 512 frames at once.
-    then __call__ returns the start of the first segment, and end (or middle, which means no end) of the last segment.
     """
 
     def reset_states(self):
@@ -137,27 +275,20 @@ class FixedVADIterator(VADIterator):
                 ret = r
             elif r is not None:
                 if "end" in r:
-                    ret["end"] = r["end"]  # the latter end
+                    ret["end"] = r["end"]
-                if "start" in r and "end" in ret:  # there is an earlier start.
+                if "start" in r and "end" in ret:
-                    # Remove end, merging this segment with the previous one.
                     del ret["end"]
         return ret if ret != {} else None
 
 
 if __name__ == "__main__":
-    # test/demonstrate the need for FixedVADIterator:
+    model = load_silero_vad(onnx=False)
-
+    vad = FixedVADIterator(model)
-    import torch
+    
-
+    audio_buffer = np.array([0] * 512, dtype=np.float32)
-    model, _ = torch.hub.load(repo_or_dir="snakers4/silero-vad", model="silero_vad")
+    result = vad(audio_buffer)
-    vac = FixedVADIterator(model)
+    print(f"   512 samples: {result}")
-    #   vac = VADIterator(model)  # the second case crashes with this
+    
-
+    # test with 511 samples
-    # this works: for both
+    audio_buffer = np.array([0] * 511, dtype=np.float32)
-    audio_buffer = np.array([0] * (512), dtype=np.float32)
+    result = vad(audio_buffer)
-    vac(audio_buffer)
-
-    # this crashes on the non FixedVADIterator with
-    # ops.prim.RaiseException("Input audio chunk is too short", "builtins.ValueError")
-    audio_buffer = np.array([0] * (512 - 1), dtype=np.float32)
-    vac(audio_buffer)

whisperlivekit/simul_whisper/backend.py

@@ -10,6 +10,7 @@ from .whisper import load_model, tokenizer
 from .whisper.audio import TOKENS_PER_SECOND
 import os
 import gc
+from pathlib import Path
 logger = logging.getLogger(__name__)
 
 import torch
@@ -22,9 +23,7 @@ try:
     HAS_MLX_WHISPER = True
 except ImportError:
     if platform.system() == "Darwin" and platform.machine() == "arm64":
-        print(f"""{"="*50}
+        print(f"""{"="*50}\nMLX Whisper not found but you are on Apple Silicon. Consider installing mlx-whisper for better performance: pip install mlx-whisper\n{"="*50}""")
-MLX Whisper not found but you are on Apple Silicon. Consider installing mlx-whisper for better performance: pip install mlx-whisper
-{"="*50}""")
     HAS_MLX_WHISPER = False
 if HAS_MLX_WHISPER:
     HAS_FASTER_WHISPER = False
@@ -35,8 +34,24 @@ else:
     except ImportError:
         HAS_FASTER_WHISPER = False
 
+def model_path_and_type(model_path):
+    path = Path(model_path)
+    
+    compatible_whisper_mlx = False
+    compatible_faster_whisper = False
+    pt_path = path if path.is_file() and path.suffix.lower() == '.pt' else None
+    
+    if path.is_dir():
+        for file in path.iterdir():
+            if file.is_file():
+                if file.name in ['weights.npz', "weights.safetensors"]:
+                    compatible_whisper_mlx = True
+                elif file.suffix.lower() == '.bin':
+                    compatible_faster_whisper = True
+                elif file.suffix.lower() == '.pt':
+                    pt_path = file
+    return pt_path, compatible_whisper_mlx, compatible_faster_whisper
 
-# TOO_MANY_REPETITIONS = 3
 
 class SimulStreamingOnlineProcessor:
     SAMPLING_RATE = 16000
@@ -154,8 +169,11 @@ class SimulStreamingASR():
             self.decoder_type = 'greedy' if self.beams == 1 else 'beam'
 
         self.fast_encoder = False
-        if self.model_dir is not None:
+        
-            self.model_path = self.model_dir
+        pt_path, compatible_whisper_mlx, compatible_faster_whisper = None, True, True
+        if self.model_path:
+            pt_path, compatible_whisper_mlx, compatible_faster_whisper = model_path_and_type(self.model_path)
+            
         elif self.model_size is not None:
             model_mapping = {
                 'tiny': './tiny.pt',
@@ -171,10 +189,12 @@ class SimulStreamingASR():
                 'large-v3': './large-v3.pt',
                 'large': './large-v3.pt'
             }
-            self.model_path = model_mapping.get(self.model_size, f'./{self.model_size}.pt')
+            pt_path = Path(model_mapping.get(self.model_size, f'./{self.model_size}.pt'))
+        
+        self.model_name = pt_path.name.replace(".pt", "")
         
         self.cfg = AlignAttConfig(
-                model_path=self.model_path,
+                tokenizer_is_multilingual= not self.model_name.endswith(".en"),
                 segment_length=self.min_chunk_size,
                 frame_threshold=self.frame_threshold,
                 language=self.lan,
@@ -196,24 +216,27 @@ class SimulStreamingASR():
         else:
             self.tokenizer = None
         
-        if self.model_dir:
+        
-            self.model_name = self.model_dir
+            
-            self.model_path = None
-        else:
-            self.model_name = os.path.basename(self.cfg.model_path).replace(".pt", "")
-            self.model_path = os.path.dirname(os.path.abspath(self.cfg.model_path))
     
         self.mlx_encoder, self.fw_encoder = None, None
         if not self.disable_fast_encoder:
             if HAS_MLX_WHISPER:
-                print('Simulstreaming will use MLX whisper for a faster encoder.')
+                print('Simulstreaming will use MLX whisper to increase encoding speed.')
-                mlx_model_name = mlx_model_mapping[self.model_name]
+                if self.model_path and compatible_whisper_mlx:
-                self.mlx_encoder = load_mlx_encoder(path_or_hf_repo=mlx_model_name)
+                    mlx_model = self.model_path
+                else:
+                    mlx_model = mlx_model_mapping[self.model_name]
+                self.mlx_encoder = load_mlx_encoder(path_or_hf_repo=mlx_model)
                 self.fast_encoder = True
-            elif HAS_FASTER_WHISPER:
+            elif HAS_FASTER_WHISPER and compatible_faster_whisper:
                 print('Simulstreaming will use Faster Whisper for the encoder.')
+                if self.model_path and compatible_faster_whisper:
+                    fw_model = self.model_path
+                else:
+                    fw_model = self.model_name
                 self.fw_encoder = WhisperModel(
-                    self.model_name,
+                    fw_model,
                     device='auto',
                     compute_type='auto',
                 )
@@ -224,7 +247,7 @@ class SimulStreamingASR():
 
     def load_model(self):
         whisper_model = load_model(
-            name=self.model_name,
+            name=self.model_path if self.model_path else self.model_name,
             download_root=self.model_path,
             decoder_only=self.fast_encoder,
             custom_alignment_heads=self.custom_alignment_heads

whisperlivekit/simul_whisper/config.py

@@ -4,26 +4,22 @@ from dataclasses import dataclass, field
 from typing import Literal
 
 @dataclass
-class SimulWhisperConfig:
+class AlignAttConfig():
-    '''Options that are common for all simul policies that could be implemented in SimulWhisper.'''
-    model_path: str
-    language: str = field(default="zh")
-    nonspeech_prob: float = 0.5
-    audio_min_len: float = 1.0
-    decoder_type: Literal["greedy","beam"] = "greedy"
-    beam_size: int = 5
-    task: Literal["transcribe","translate"] = "transcribe"
-    init_prompt: str = field(default=None)
-    static_init_prompt: str = field(default=None)
-    max_context_tokens: int = field(default=None)
-
-@dataclass
-class AlignAttConfig(SimulWhisperConfig):
-    '''Options specific to the AlignAtt policy.'''
     eval_data_path: str = "tmp"
     segment_length: float = field(default=1.0, metadata = {"help": "in second"})
     frame_threshold: int = 4
     rewind_threshold: int = 200
     audio_max_len: float = 20.0
     cif_ckpt_path: str = ""
-    never_fire: bool = False
+    never_fire: bool = False
+    language: str = field(default="zh")
+    nonspeech_prob: float = 0.5
+    audio_min_len: float = 1.0
+    decoder_type: Literal["greedy","beam"] = "greedy"
+    beam_size: int = 5
+    task: Literal["transcribe","translate"] = "transcribe"
+    tokenizer_is_multilingual: bool = False
+    init_prompt: str = field(default=None)
+    static_init_prompt: str = field(default=None)
+    max_context_tokens: int = field(default=None)
+    

whisperlivekit/simul_whisper/license_simulstreaming.py

@@ -1,5 +0,0 @@
-SIMULSTREAMING_LICENSE = f"""
-SimulStreaming backend is dual-licensed:
-• Non-Commercial Use: PolyForm Noncommercial License 1.0.0.
-• Commercial Use: Check SimulStreaming README (github.com/ufal/SimulStreaming) for more details.
-"""

whisperlivekit/simul_whisper/simul_whisper.py

@@ -51,20 +51,15 @@ class PaddedAlignAttWhisper:
             fw_encoder=None,
         ) -> None:
         self.log_segments = 0
-        model_name = os.path.basename(cfg.model_path).replace(".pt", "")
-        model_path = os.path.dirname(os.path.abspath(cfg.model_path))
-        if loaded_model:
-            self.model = loaded_model
-        else:
-            self.model = load_model(name=model_name, download_root=model_path)
         
-        self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
+        self.model = loaded_model
-
         self.mlx_encoder = mlx_encoder
         self.fw_encoder = fw_encoder
         if fw_encoder:
             self.fw_feature_extractor = FeatureExtractor(feature_size=self.model.dims.n_mels)
-            
+        
+        self.device = 'cuda' if torch.cuda.is_available() else 'cpu'
+        
         logger.info(f"Model dimensions: {self.model.dims}")
         self.speaker = -1
         self.decode_options = DecodingOptions(
@@ -72,7 +67,7 @@ class PaddedAlignAttWhisper:
             without_timestamps = True,
             task=cfg.task
         )
-        self.tokenizer_is_multilingual = not model_name.endswith(".en")
+        self.tokenizer_is_multilingual = cfg.tokenizer_is_multilingual
         self.create_tokenizer(cfg.language if cfg.language != "auto" else None)
         # self.create_tokenizer('en')
         self.detected_language = cfg.language if cfg.language != "auto" else None
@@ -172,7 +167,10 @@ class PaddedAlignAttWhisper:
             self.inference.kv_cache = self.kv_cache
 
             self.token_decoder = BeamSearchDecoder(inference=self.inference, eot=self.tokenizer.eot, beam_size=cfg.beam_size)
-            
+
+        # Tokens to carry over to next chunk for incomplete UTF-8 characters
+        self.pending_incomplete_tokens = []
+
     def remove_hooks(self):
         for hook in self.l_hooks:
             hook.remove()
@@ -266,6 +264,7 @@ class PaddedAlignAttWhisper:
             self.segments = []
         self.log_segments += 1
 
+        self.pending_incomplete_tokens = []
 
     def fire_at_boundary(self, chunked_encoder_feature: torch.Tensor):
         if self.always_fire: return True
@@ -327,7 +326,7 @@ class PaddedAlignAttWhisper:
             self.segments = self.segments[1:]
             logger.debug(f"remove segments: {len(self.segments)} {len(self.tokens)}, cumulative offset: {self.cumulative_time_offset:.2f}s")
             if len(self.tokens) > 1:
-                self.context.append_token_ids(self.tokens[1][0,:])
+                self.context.append_token_ids(self.tokens[1][0,:].tolist())
                 self.tokens = [self.initial_tokens] + self.tokens[2:]
         return removed_len
 
@@ -567,6 +566,12 @@ class PaddedAlignAttWhisper:
 
         tokens_to_split = current_tokens[0, token_len_before_decoding:]
 
+        # Prepend pending tokens from previous chunk if any
+        if self.pending_incomplete_tokens:
+            logger.debug(f"[UTF-8 Fix] Prepending {len(self.pending_incomplete_tokens)} pending tokens: {self.pending_incomplete_tokens}")
+            pending_tensor = torch.tensor(self.pending_incomplete_tokens, dtype=torch.long, device=self.device)
+            tokens_to_split = torch.cat([pending_tensor, tokens_to_split])
+
         if fire_detected or is_last: #or punctuation_stop:
             new_hypothesis = tokens_to_split.flatten().tolist()
             split_words, split_tokens = self.tokenizer.split_to_word_tokens(new_hypothesis)
@@ -595,7 +600,14 @@ class PaddedAlignAttWhisper:
 
         timestamped_words = []
         timestamp_idx = 0
+        replacement_char = "\ufffd"
         for word, word_tokens in zip(split_words, split_tokens):
+            # Skip words containing incomplete UTF-8 from client output
+            if replacement_char in word:
+                logger.warning(f"[UTF-8 Filter] Skipping incomplete word from client output: {repr(word)}")
+                timestamp_idx += len(word_tokens)
+                continue
+
             try:
                 current_timestamp = l_absolute_timestamps[timestamp_idx]
             except:
@@ -613,5 +625,11 @@ class PaddedAlignAttWhisper:
                     self.global_time_offset
             )
             timestamped_words.append(timestamp_entry)
-        
+
-        return timestamped_words
+        # Hold incomplete tokens for next chunk
+        self.pending_incomplete_tokens = []
+        if split_words and replacement_char in split_words[-1]:
+            self.pending_incomplete_tokens = split_tokens[-1]
+            logger.warning(f"[UTF-8 Fix] Holding {len(self.pending_incomplete_tokens)} incomplete tokens for next chunk: {self.pending_incomplete_tokens}")
+
+        return timestamped_words

whisperlivekit/simul_whisper/token_buffer.py

@@ -7,6 +7,7 @@ class TokenBuffer:
         self.prefix_token_ids = prefix_token_ids
         self.tokenizer = tokenizer
         self.device = device
+        self.pending_token_ids = []
 
     def as_token_ids(self, tokenizer=None):
 
@@ -64,7 +65,26 @@ class TokenBuffer:
     def append_token_ids(self, token_ids):
         tokenizer = self.tokenizer
         assert tokenizer is not None, "Tokenizer is not set."
-        self.text += self.tokenizer.decode(token_ids)
+
+        all_tokens = self.pending_token_ids + token_ids
+
+        decoded = tokenizer.decode(all_tokens)
+        replacement_char = "\ufffd"
+
+        if replacement_char in decoded:
+            if len(all_tokens) > 1:
+                decoded_partial = tokenizer.decode(all_tokens[:-1])
+
+                if replacement_char not in decoded_partial:
+                    self.text += decoded_partial
+                    self.pending_token_ids = [all_tokens[-1]]
+                else:
+                    self.pending_token_ids = all_tokens
+            else:
+                self.pending_token_ids = all_tokens
+        else:
+            self.text += decoded
+            self.pending_token_ids = []
 
     def as_split_word_tokens(self):
         tokenizer = self.tokenizer

whisperlivekit/timed_objects.py

@@ -174,11 +174,13 @@ class ChangeSpeaker:
 
 @dataclass  
 class State():
-    tokens: list
+    tokens: list = field(default_factory=list)
-    last_validated_token: int
+    last_validated_token: int = 0
-    translated_segments: list
+    translation_validated_segments: list = field(default_factory=list)
-    buffer_transcription: str
+    translation_buffer: list = field(default_factory=list)
-    end_buffer: float
+    buffer_transcription: str = field(default_factory=Transcript)
-    end_attributed_speaker: float
+    end_buffer: float = 0.0
-    remaining_time_transcription: float
+    end_attributed_speaker: float = 0.0
-    remaining_time_diarization: float
+    remaining_time_transcription: float = 0.0
+    remaining_time_diarization: float = 0.0
+    beg_loop: Optional[int] = None

whisperlivekit/translation/mapping_languages.py

@@ -1,182 +0,0 @@
-"""
-adapted from https://store.crowdin.com/custom-mt
-"""
-
-LANGUAGES = [
-    {"name": "Afrikaans", "nllb": "afr_Latn", "crowdin": "af"},
-    {"name": "Akan", "nllb": "aka_Latn", "crowdin": "ak"},
-    {"name": "Amharic", "nllb": "amh_Ethi", "crowdin": "am"},
-    {"name": "Assamese", "nllb": "asm_Beng", "crowdin": "as"},
-    {"name": "Asturian", "nllb": "ast_Latn", "crowdin": "ast"},
-    {"name": "Bashkir", "nllb": "bak_Cyrl", "crowdin": "ba"},
-    {"name": "Bambara", "nllb": "bam_Latn", "crowdin": "bm"},
-    {"name": "Balinese", "nllb": "ban_Latn", "crowdin": "ban"},
-    {"name": "Belarusian", "nllb": "bel_Cyrl", "crowdin": "be"},
-    {"name": "Bengali", "nllb": "ben_Beng", "crowdin": "bn"},
-    {"name": "Bosnian", "nllb": "bos_Latn", "crowdin": "bs"},
-    {"name": "Bulgarian", "nllb": "bul_Cyrl", "crowdin": "bg"},
-    {"name": "Catalan", "nllb": "cat_Latn", "crowdin": "ca"},
-    {"name": "Cebuano", "nllb": "ceb_Latn", "crowdin": "ceb"},
-    {"name": "Czech", "nllb": "ces_Latn", "crowdin": "cs"},
-    {"name": "Welsh", "nllb": "cym_Latn", "crowdin": "cy"},
-    {"name": "Danish", "nllb": "dan_Latn", "crowdin": "da"},
-    {"name": "German", "nllb": "deu_Latn", "crowdin": "de"},
-    {"name": "Dzongkha", "nllb": "dzo_Tibt", "crowdin": "dz"},
-    {"name": "Greek", "nllb": "ell_Grek", "crowdin": "el"},
-    {"name": "English", "nllb": "eng_Latn", "crowdin": "en"},
-    {"name": "Esperanto", "nllb": "epo_Latn", "crowdin": "eo"},
-    {"name": "Estonian", "nllb": "est_Latn", "crowdin": "et"},
-    {"name": "Basque", "nllb": "eus_Latn", "crowdin": "eu"},
-    {"name": "Ewe", "nllb": "ewe_Latn", "crowdin": "ee"},
-    {"name": "Faroese", "nllb": "fao_Latn", "crowdin": "fo"},
-    {"name": "Fijian", "nllb": "fij_Latn", "crowdin": "fj"},
-    {"name": "Finnish", "nllb": "fin_Latn", "crowdin": "fi"},
-    {"name": "French", "nllb": "fra_Latn", "crowdin": "fr"},
-    {"name": "Friulian", "nllb": "fur_Latn", "crowdin": "fur-IT"},
-    {"name": "Scottish Gaelic", "nllb": "gla_Latn", "crowdin": "gd"},
-    {"name": "Irish", "nllb": "gle_Latn", "crowdin": "ga-IE"},
-    {"name": "Galician", "nllb": "glg_Latn", "crowdin": "gl"},
-    {"name": "Guarani", "nllb": "grn_Latn", "crowdin": "gn"},
-    {"name": "Gujarati", "nllb": "guj_Gujr", "crowdin": "gu-IN"},
-    {"name": "Haitian Creole", "nllb": "hat_Latn", "crowdin": "ht"},
-    {"name": "Hausa", "nllb": "hau_Latn", "crowdin": "ha"},
-    {"name": "Hebrew", "nllb": "heb_Hebr", "crowdin": "he"},
-    {"name": "Hindi", "nllb": "hin_Deva", "crowdin": "hi"},
-    {"name": "Croatian", "nllb": "hrv_Latn", "crowdin": "hr"},
-    {"name": "Hungarian", "nllb": "hun_Latn", "crowdin": "hu"},
-    {"name": "Armenian", "nllb": "hye_Armn", "crowdin": "hy-AM"},
-    {"name": "Igbo", "nllb": "ibo_Latn", "crowdin": "ig"},
-    {"name": "Indonesian", "nllb": "ind_Latn", "crowdin": "id"},
-    {"name": "Icelandic", "nllb": "isl_Latn", "crowdin": "is"},
-    {"name": "Italian", "nllb": "ita_Latn", "crowdin": "it"},
-    {"name": "Javanese", "nllb": "jav_Latn", "crowdin": "jv"},
-    {"name": "Japanese", "nllb": "jpn_Jpan", "crowdin": "ja"},
-    {"name": "Kabyle", "nllb": "kab_Latn", "crowdin": "kab"},
-    {"name": "Kannada", "nllb": "kan_Knda", "crowdin": "kn"},
-    {"name": "Georgian", "nllb": "kat_Geor", "crowdin": "ka"},
-    {"name": "Kazakh", "nllb": "kaz_Cyrl", "crowdin": "kk"},
-    {"name": "Khmer", "nllb": "khm_Khmr", "crowdin": "km"},
-    {"name": "Kinyarwanda", "nllb": "kin_Latn", "crowdin": "rw"},
-    {"name": "Kyrgyz", "nllb": "kir_Cyrl", "crowdin": "ky"},
-    {"name": "Korean", "nllb": "kor_Hang", "crowdin": "ko"},
-    {"name": "Lao", "nllb": "lao_Laoo", "crowdin": "lo"},
-    {"name": "Ligurian", "nllb": "lij_Latn", "crowdin": "lij"},
-    {"name": "Limburgish", "nllb": "lim_Latn", "crowdin": "li"},
-    {"name": "Lingala", "nllb": "lin_Latn", "crowdin": "ln"},
-    {"name": "Lithuanian", "nllb": "lit_Latn", "crowdin": "lt"},
-    {"name": "Luxembourgish", "nllb": "ltz_Latn", "crowdin": "lb"},
-    {"name": "Maithili", "nllb": "mai_Deva", "crowdin": "mai"},
-    {"name": "Malayalam", "nllb": "mal_Mlym", "crowdin": "ml-IN"},
-    {"name": "Marathi", "nllb": "mar_Deva", "crowdin": "mr"},
-    {"name": "Macedonian", "nllb": "mkd_Cyrl", "crowdin": "mk"},
-    {"name": "Maltese", "nllb": "mlt_Latn", "crowdin": "mt"},
-    {"name": "Mossi", "nllb": "mos_Latn", "crowdin": "mos"},
-    {"name": "Maori", "nllb": "mri_Latn", "crowdin": "mi"},
-    {"name": "Burmese", "nllb": "mya_Mymr", "crowdin": "my"},
-    {"name": "Dutch", "nllb": "nld_Latn", "crowdin": "nl"},
-    {"name": "Norwegian Nynorsk", "nllb": "nno_Latn", "crowdin": "nn-NO"},
-    {"name": "Nepali", "nllb": "npi_Deva", "crowdin": "ne-NP"},
-    {"name": "Northern Sotho", "nllb": "nso_Latn", "crowdin": "nso"},
-    {"name": "Occitan", "nllb": "oci_Latn", "crowdin": "oc"},
-    {"name": "Odia", "nllb": "ory_Orya", "crowdin": "or"},
-    {"name": "Papiamento", "nllb": "pap_Latn", "crowdin": "pap"},
-    {"name": "Polish", "nllb": "pol_Latn", "crowdin": "pl"},
-    {"name": "Portuguese", "nllb": "por_Latn", "crowdin": "pt-PT"},
-    {"name": "Dari", "nllb": "prs_Arab", "crowdin": "fa-AF"},
-    {"name": "Romanian", "nllb": "ron_Latn", "crowdin": "ro"},
-    {"name": "Rundi", "nllb": "run_Latn", "crowdin": "rn"},
-    {"name": "Russian", "nllb": "rus_Cyrl", "crowdin": "ru"},
-    {"name": "Sango", "nllb": "sag_Latn", "crowdin": "sg"},
-    {"name": "Sanskrit", "nllb": "san_Deva", "crowdin": "sa"},
-    {"name": "Santali", "nllb": "sat_Olck", "crowdin": "sat"},
-    {"name": "Sinhala", "nllb": "sin_Sinh", "crowdin": "si-LK"},
-    {"name": "Slovak", "nllb": "slk_Latn", "crowdin": "sk"},
-    {"name": "Slovenian", "nllb": "slv_Latn", "crowdin": "sl"},
-    {"name": "Shona", "nllb": "sna_Latn", "crowdin": "sn"},
-    {"name": "Sindhi", "nllb": "snd_Arab", "crowdin": "sd"},
-    {"name": "Somali", "nllb": "som_Latn", "crowdin": "so"},
-    {"name": "Southern Sotho", "nllb": "sot_Latn", "crowdin": "st"},
-    {"name": "Spanish", "nllb": "spa_Latn", "crowdin": "es-ES"},
-    {"name": "Sardinian", "nllb": "srd_Latn", "crowdin": "sc"},
-    {"name": "Swati", "nllb": "ssw_Latn", "crowdin": "ss"},
-    {"name": "Sundanese", "nllb": "sun_Latn", "crowdin": "su"},
-    {"name": "Swedish", "nllb": "swe_Latn", "crowdin": "sv-SE"},
-    {"name": "Swahili", "nllb": "swh_Latn", "crowdin": "sw"},
-    {"name": "Tamil", "nllb": "tam_Taml", "crowdin": "ta"},
-    {"name": "Tatar", "nllb": "tat_Cyrl", "crowdin": "tt-RU"},
-    {"name": "Telugu", "nllb": "tel_Telu", "crowdin": "te"},
-    {"name": "Tajik", "nllb": "tgk_Cyrl", "crowdin": "tg"},
-    {"name": "Tagalog", "nllb": "tgl_Latn", "crowdin": "tl"},
-    {"name": "Thai", "nllb": "tha_Thai", "crowdin": "th"},
-    {"name": "Tigrinya", "nllb": "tir_Ethi", "crowdin": "ti"},
-    {"name": "Tswana", "nllb": "tsn_Latn", "crowdin": "tn"},
-    {"name": "Tsonga", "nllb": "tso_Latn", "crowdin": "ts"},
-    {"name": "Turkmen", "nllb": "tuk_Latn", "crowdin": "tk"},
-    {"name": "Turkish", "nllb": "tur_Latn", "crowdin": "tr"},
-    {"name": "Uyghur", "nllb": "uig_Arab", "crowdin": "ug"},
-    {"name": "Ukrainian", "nllb": "ukr_Cyrl", "crowdin": "uk"},
-    {"name": "Venetian", "nllb": "vec_Latn", "crowdin": "vec"},
-    {"name": "Vietnamese", "nllb": "vie_Latn", "crowdin": "vi"},
-    {"name": "Wolof", "nllb": "wol_Latn", "crowdin": "wo"},
-    {"name": "Xhosa", "nllb": "xho_Latn", "crowdin": "xh"},
-    {"name": "Yoruba", "nllb": "yor_Latn", "crowdin": "yo"},
-    {"name": "Zulu", "nllb": "zul_Latn", "crowdin": "zu"},
-]
-
-NAME_TO_NLLB = {lang["name"]: lang["nllb"] for lang in LANGUAGES}
-NAME_TO_CROWDIN = {lang["name"]: lang["crowdin"] for lang in LANGUAGES}
-CROWDIN_TO_NLLB = {lang["crowdin"]: lang["nllb"] for lang in LANGUAGES}
-NLLB_TO_CROWDIN = {lang["nllb"]: lang["crowdin"] for lang in LANGUAGES}
-CROWDIN_TO_NAME = {lang["crowdin"]: lang["name"] for lang in LANGUAGES}
-NLLB_TO_NAME = {lang["nllb"]: lang["name"] for lang in LANGUAGES}
-
-
-def get_nllb_code(crowdin_code):
-    return CROWDIN_TO_NLLB.get(crowdin_code, None)
-
-
-def get_crowdin_code(nllb_code):
-    return NLLB_TO_CROWDIN.get(nllb_code)
-
-
-def get_language_name_by_crowdin(crowdin_code):
-    return CROWDIN_TO_NAME.get(crowdin_code)
-
-
-def get_language_name_by_nllb(nllb_code):
-    return NLLB_TO_NAME.get(nllb_code)
-
-
-def get_language_info(identifier, identifier_type="auto"):
-    if identifier_type == "auto":
-        for lang in LANGUAGES:
-            if (lang["name"].lower() == identifier.lower() or 
-                lang["nllb"] == identifier or 
-                lang["crowdin"] == identifier):
-                return lang
-    elif identifier_type == "name":
-        for lang in LANGUAGES:
-            if lang["name"].lower() == identifier.lower():
-                return lang
-    elif identifier_type == "nllb":
-        for lang in LANGUAGES:
-            if lang["nllb"] == identifier:
-                return lang
-    elif identifier_type == "crowdin":
-        for lang in LANGUAGES:
-            if lang["crowdin"] == identifier:
-                return lang
-    
-    return None
-
-
-def list_all_languages():
-    return [lang["name"] for lang in LANGUAGES]
-
-
-def list_all_nllb_codes():
-    return [lang["nllb"] for lang in LANGUAGES]
-
-
-def list_all_crowdin_codes():
-    return [lang["crowdin"] for lang in LANGUAGES]

whisperlivekit/translation/translation.py

@@ -1,169 +0,0 @@
-import logging
-import time
-import ctranslate2
-import torch
-import transformers
-from dataclasses import dataclass, field
-import huggingface_hub
-from whisperlivekit.translation.mapping_languages import get_nllb_code
-from whisperlivekit.timed_objects import Translation
-
-logger = logging.getLogger(__name__)
-
-#In diarization case, we may want to translate just one speaker, or at least start the sentences there
-
-MIN_SILENCE_DURATION_DEL_BUFFER = 3 #After a silence of x seconds, we consider the model should not use the buffer, even if the previous
-# sentence is not finished.
-
-@dataclass
-class TranslationModel():
-    translator: ctranslate2.Translator
-    device: str
-    tokenizer: dict = field(default_factory=dict)
-    backend_type: str = 'ctranslate2'
-    nllb_size: str = '600M'
-    
-    def get_tokenizer(self, input_lang):
-        if not self.tokenizer.get(input_lang, False):
-            self.tokenizer[input_lang] = transformers.AutoTokenizer.from_pretrained(
-                f"facebook/nllb-200-distilled-{self.nllb_size}",
-                src_lang=input_lang,
-                clean_up_tokenization_spaces=True
-            )
-        return self.tokenizer[input_lang]
-            
-
-def load_model(src_langs, nllb_backend='ctranslate2', nllb_size='600M'):
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-    MODEL = f'nllb-200-distilled-{nllb_size}-ctranslate2'
-    if nllb_backend=='ctranslate2':
-        MODEL_GUY = 'entai2965'
-        huggingface_hub.snapshot_download(MODEL_GUY + '/' + MODEL,local_dir=MODEL)
-        translator = ctranslate2.Translator(MODEL,device=device)
-    elif nllb_backend=='transformers':
-        translator = transformers.AutoModelForSeq2SeqLM.from_pretrained(f"facebook/nllb-200-distilled-{nllb_size}")
-    tokenizer = dict()
-    for src_lang in src_langs:
-        if src_lang != 'auto':
-            tokenizer[src_lang] = transformers.AutoTokenizer.from_pretrained(MODEL, src_lang=src_lang, clean_up_tokenization_spaces=True)
-
-    translation_model = TranslationModel(
-        translator=translator,
-        tokenizer=tokenizer,
-        backend_type=nllb_backend,
-        device = device,
-        nllb_size = nllb_size
-    )
-    for src_lang in src_langs:
-        if src_lang != 'auto':
-            translation_model.get_tokenizer(src_lang)
-    return translation_model
-
-class OnlineTranslation:
-    def __init__(self, translation_model: TranslationModel, input_languages: list, output_languages: list):
-        self.buffer = []
-        self.len_processed_buffer = 0
-        self.translation_remaining = Translation()
-        self.validated = []
-        self.translation_pending_validation = ''
-        self.translation_model = translation_model
-        self.input_languages = input_languages
-        self.output_languages = output_languages
-
-    def compute_common_prefix(self, results):
-        #we dont want want to prune the result for the moment. 
-        if not self.buffer:
-            self.buffer = results
-        else:
-            for i in range(min(len(self.buffer), len(results))):
-                if self.buffer[i] != results[i]:
-                    self.commited.extend(self.buffer[:i])
-                    self.buffer = results[i:]
-
-    def translate(self, input, input_lang, output_lang):
-        if not input:
-            return ""
-        nllb_output_lang = get_nllb_code(output_lang)
-            
-        tokenizer = self.translation_model.get_tokenizer(input_lang)
-        tokenizer_output = tokenizer(input, return_tensors="pt").to(self.translation_model.device)
-        
-        if self.translation_model.backend_type == 'ctranslate2':
-            source = tokenizer.convert_ids_to_tokens(tokenizer_output['input_ids'][0])    
-            results = self.translation_model.translator.translate_batch([source], target_prefix=[[nllb_output_lang]])
-            target = results[0].hypotheses[0][1:]
-            result = tokenizer.decode(tokenizer.convert_tokens_to_ids(target))
-        else:
-            translated_tokens = self.translation_model.translator.generate(**tokenizer_output, forced_bos_token_id=tokenizer.convert_tokens_to_ids(nllb_output_lang))
-            result = tokenizer.batch_decode(translated_tokens, skip_special_tokens=True)[0]
-        return result
-    
-    def translate_tokens(self, tokens):
-        if tokens:
-            text = ' '.join([token.text for token in tokens])
-            start = tokens[0].start
-            end = tokens[-1].end
-            if self.input_languages[0] == 'auto':
-                input_lang = tokens[0].detected_language
-            else:
-                input_lang = self.input_languages[0]
-                
-            translated_text = self.translate(text,
-                                            input_lang,
-                                            self.output_languages[0]
-                                            )
-            translation = Translation(
-                text=translated_text,
-                start=start,
-                end=end,
-            )
-            return translation
-        return None
-            
-
-    def insert_tokens(self, tokens):
-        self.buffer.extend(tokens)
-        pass
-    
-    def process(self):
-        i = 0
-        if len(self.buffer) < self.len_processed_buffer + 3: #nothing new to process
-            return self.validated + [self.translation_remaining]
-        while i < len(self.buffer):
-            if self.buffer[i].is_punctuation():
-                translation_sentence = self.translate_tokens(self.buffer[:i+1])
-                self.validated.append(translation_sentence)
-                self.buffer = self.buffer[i+1:]
-                i = 0
-            else:
-                i+=1
-        self.translation_remaining = self.translate_tokens(self.buffer)
-        self.len_processed_buffer = len(self.buffer)
-        return self.validated + [self.translation_remaining]
-
-    def insert_silence(self, silence_duration: float):
-        if silence_duration >= MIN_SILENCE_DURATION_DEL_BUFFER:
-            self.buffer = []
-            self.validated += [self.translation_remaining]
-
-if __name__ == '__main__':
-    output_lang = 'fr'
-    input_lang = "en"
-    
-    
-    test_string = """
-    Transcription technology has improved so much in the past few years. Have you noticed how accurate real-time speech-to-text is now?
-    """
-    test = test_string.split(' ')
-    step = len(test) // 3
-    
-    shared_model = load_model([input_lang], nllb_backend='ctranslate2')
-    online_translation = OnlineTranslation(shared_model, input_languages=[input_lang], output_languages=[output_lang])
-    
-    beg_inference = time.time()    
-    for id in range(5):
-        val = test[id*step : (id+1)*step]
-        val_str = ' '.join(val)
-        result = online_translation.translate(val_str)
-        print(result)
-    print('inference time:', time.time() - beg_inference)