Speech Recognition Channels
Speech recognition (ASR) is the first step in video translation. It converts the spoken audio or video into subtitle files with timestamps. pyVideoTrans supports 15+ recognition channels.
If you are not sure what language is spoken in the video, you can use the
Speech-To-Textfeature in the left panel. Set the spoken language toAuto Detectand choose a recognition channel with broad language support, such asfaster-whisper/openai-whisper/Omnilingual/Dolphin.
Local Offline Recognition
Works completely offline without an internet connection. Models are downloaded automatically upon first use.
| Channel | Description | GPU Acceleration | Recommendation |
|---|---|---|---|
| faster-whisper(Built-in) | Fast speed and high quality, supports all built-in languages | ✅ | ⭐⭐⭐ Default Recommendation |
| openai-whisper(Built-in) | High accuracy, slightly slower, supports all built-in languages | ✅ | ⭐⭐⭐ |
| Qwen-ASR(Built-in) | Excellent for Chinese, supports most built-in languages | ✅ | ⭐⭐⭐ |
Whisper.cpp(Win Built-in) | Built into Windows for direct use; requires separate manual deployment on macOS and Linux | ✅ | ⭐⭐⭐ |
| FunASR(Built-in) | Excellent for Chinese, with multiple models available | ✅ | ⭐⭐⭐ |
| Firered Chinese(Built-in) | Only supports Chinese and 20 Chinese dialects | X | ⭐⭐ |
| Dolphin(Built-in) | Supports 40+ Asian languages and 20 Chinese dialects | X | ⭐⭐ |
| Omnilingual ASR(Built-in) | Supports all built-in languages and 1,600+ more | X | ⭐⭐ |
| Nemotron-3.5-asr-0.6(Built-in) | English,japanese,ko,vietnamese,eg. 40 | ✅ | ⭐⭐ |
| Huggingface_ASR(Built-in) | Multiple language models available to choose from | ✅ | ⭐⭐ |
| Moss-Diarize(Built-in) | Supports 50+ languages; can separate speakers for audio under 90 minutes. Files over 90 minutes require a separate speaker model | ✅ | ⭐⭐ |
| Faster-Whisper-XXL.exe | A standalone Windows package of faster-whisper. Requires manual download and specifying the .exe path | ✅ | ⭐⭐ |
Because models are quite large and network environments vary, automatic downloads might sometimes fail. If a download fails, click here to view model download links and manual setup instructions.
Model Selection Guide for faster-whisper / openai-whisper
| Model | Speed | Accuracy | VRAM Required |
|---|---|---|---|
| tiny | Fastest | Low | ~1GB |
| base | Fast | Low-Medium | ~1GB |
| small | Medium | Medium | ~2GB |
| medium | Slow | High | ~5GB |
| large-v3 | Slowest | Highest | ~8GB |
| large-v3-turbo | Fast | High | ~6GB |
Recommendation: large-v3-turbo offers the best balance between speed and quality.
Online Recognition
| Channel | Description |
|---|---|
| Alibaba Cloud Bailian Qwen3-ASR | Requires activating the Alibaba Cloud Bailian service |
| Xiaomi | The mimo-v2.5-asr model works great for Chinese and mixed Chinese-English. Requires activating the Xiaomi AI platform, topping up, and getting an API key. Enter it under Menu -> Translation Settings -> Xiaomi AI |
| ByteDance Speech Recognition Large Model Express Edition | Outstanding performance on Chinese |
| Elevenlabs.io Speech Recognition | Free accounts have strict rate limits and are barely usable |
| Deepgram.com | Requires registering for an API Key |
| Gemini AI | Strong at recognizing low-resource/minority languages; requires proxy/VPN access in restricted regions |
| 302.AI | Apply on 302.ai |
| OpenAI Speech Recognition API | Excellent quality, requires an OpenAI API key (SK key) |
Advanced & Custom
| Channel | Description |
|---|---|
| Parakeet-tdt(LocalAPI) | Requires separate manual deployment |
| WhisperX(LocalAPI) | Requires separate manual deployment |
| STT(LocalAPI) | Requires separate manual deployment |
| Whisper.NET | Supports AMD GPU acceleration. Requires source code installation and downloading the required DLL files according to the guide |
| Custom Speech Recognition API | Allows you to connect your own custom speech recognition API endpoint |
Available Models for Huggingface_ASR(Built-in)
| Model | Supported Languages |
|---|---|
| nvidia/parakeet-tdt-0.6b-v3 | en,bg,hr,cs,da,nl,et,fi,fr,de,el,hu,it,lv,lt,mt,pl,pt,ro,sk,sl,es,sv,ru,uk |
| nvidia/nemotron-3.5-asr-streaming-0.6b | en,bg,hr,cs,ko,ja,vi,.eg 40 |
| reazon-research/japanese-wav2vec2-large-rs35kh | Japanese |
| kotoba-tech/kotoba-whisper-v2.0 | Japanese |
| biodatlab/whisper-th-large-v3 | Thai |
| vinai/Phowhisper-large | Vietnamese |
| anke01/whisper-small-uyghur | Uyghur |
| openai/whisper-large-v3 | All languages |
| zai-org/GLM-ASR-Nano-2512 | Zhipu AI model, excellent for Mandarin and Cantonese |
| Audio8/ARK-ASR-0.6B/3B | Chinese, English, German, Japanese, French, Korean, Spanish, Polish, Italian, Romanian, Hungarian, Czech, Dutch |
| ibm-granite/granite-speech-4.1-2b | English, French, German, Spanish, Portuguese, Japanese |
| nguyenvulebinh/wav2vec2-base-vietnamese-250h | vietnamese |
| sakares/wav2vec2-large-xlsr-thai-demo | thai |
| SiangLao/xlsr-53-lao-asr | lao |
| chuuhtetnaing/whisper-large-v3-myanmar | myanmar |
| 1morecupofhottea/whisper-turbo-khmer-v9 | khmer |
| anke01/whisper-small-uyghur | uyghur |
| kingabzpro/whisper-large-v3-turbo-urdu | urdu |
| vasista22/whisper-tamil-small | tamil |
| theainerd/Wav2Vec2-large-xlsr-hindi | hindi |
| cautroi/whisper-large-v3-id | Indonesia |
| Khalsuu/filipino-wav2vec2-l-xls-r-300m-official | Filipino |
| navai-uz/whisper-medium-uzbek | uzbek |
| jonatasgrosman/wav2vec2-large-xlsr-53-persian | persian |
| Ghost3454/translynx-pakistani-punjabi-whisper-small | pakistani |
| turkmedstt/whisper-large-v3-turkish-general | turkish |
| anton-l/wav2vec2-large-xlsr-53-mongolian | mongolian |
| HNO333333/w2v-bert-2.0-Tibetan-Amdo | Tibetan |