SenseVoice: fast ASR for Chinese, Cantonese, Japanese and Korean
Don’t want to install SenseVoice? Drop the file here.
The first 20 minutes of any recording are transcribed free: a recording up to 20 minutes transcribes in full, a longer one is previewed, and a pack from $6 (500 minutes) transcribes the rest and opens every export, and credits never expire.
SenseVoiceSmall transcribes five languages, identifies which one is spoken, and tags emotion and sounds like laughter or applause, with a non-autoregressive design that is far faster than Whisper.
Get SenseVoice Official site & pricing
SenseVoiceSmall transcribes Mandarin, Cantonese, English, Japanese and Korean, detects the language, and tags emotion and audio events. Its authors measure it at 15× the speed of Whisper-Large.
Part of the FunASR toolkit. Also inside apps like Whisper Notes and VoiceInk. For Cantonese recordings with no setup, a hosted service takes the file.
What it is
SenseVoice is a speech foundation model from the FunAudioLLM team (the repository now lives at QwenAudio/SenseVoice). The released SenseVoiceSmall checkpoint does speech recognition and language identification for Mandarin, Cantonese, English, Japanese and Korean, plus speech-emotion recognition and audio-event detection (background music, applause, laughter, crying, coughing, sneezing). It is non-autoregressive; in the authors' benchmark it runs more than 5 times faster than Whisper-Small and 15 times faster than Whisper-Large, and they report an edge over Whisper on Chinese and Cantonese test sets. The broader research covers 50+ languages, but the public checkpoint does not. Code is MIT; the weights on Hugging Face are under the FunASR model licence. It runs through FunASR, which also offers a composed diarization pipeline.
Watch out for: The released SenseVoiceSmall covers five languages only. Long recordings must be segmented (FSMN-VAD or its long_audio_no_vad.py script), and speaker diarization is a separate FunASR pipeline, not a model output.
Cantonese and Mandarin recordings, hosted
- Drop the audio or video file into the box at the top of this page, or paste a link (YouTube, a podcast or RSS feed, Google Drive, Twitch or Kick). Nothing to install and no GPU needed.
- The language is detected automatically, Cantonese and Mandarin included, and the transcript stays in the language that was spoken. WhipScribe does not translate.
- Read the summary free. A pack from $6 (500 minutes) opens the full transcript and the downloads: TXT, DOCX, SRT, VTT and JSON. Packs are one-time purchases and credits never expire.
- Files can be up to 10 hours or 5 GB per file.
Related: Transcribe Cantonese · CUHK Cantonese ASR · FunASR · SRT generator · Pricing
Install / use
pip install -U "funasr>=1.3.26" # model: iic/SenseVoiceSmallKey specs
What people actually do with SenseVoice-style transcription
The tool is the means. These are the jobs — each one priced at published rates, each one wired up on its own page.
Features
| Speaker diarization | No |
| Word-level timestamps | Not stated |
| Streaming / real-time | Not stated |
| Languages supported | 5 |
| HIPAA eligible | No |
Links
- github.com/QwenAudio/SenseVoice ↗MIT code · benchmarks · long-audio script
- FunAudioLLM/SenseVoiceSmall ↗model weights · FunASR model licence
- FunAudioLLM paper ↗arXiv 2407.04051
SenseVoice vs Whipscribe
| Feature | SenseVoice | Whipscribe |
|---|---|---|
| Category | Open source | Transcription APIs |
| Pricing | free (MIT code; model under the FunASR model licence) | $6–$24 one-time packs (credits never expire) · first 20 minutes free |
| Speaker identification | No | Named in the summary, key quotes and chapters |
| Word timestamps | No | Yes |
| Streaming | No | No |
| Languages | 5 | 99 |
| Platforms | Linux, macOS, Windows | Web, API, MCP |
Alternatives to SenseVoice
Frequently asked about SenseVoice
Which languages does SenseVoice support?
The released SenseVoiceSmall checkpoint supports Mandarin, Cantonese, English, Japanese and Korean (language codes zh, yue, en, ja, ko, or auto). The 50+ languages in the research paper are not in the public checkpoint.
Is SenseVoice faster than Whisper?
In the authors' benchmark, SenseVoiceSmall runs more than 5 times faster than Whisper-Small and 15 times faster than Whisper-Large, because it is non-autoregressive.
Does SenseVoice do speaker diarization?
Not by itself. The README says diarization is a composed FunASR pipeline with separate FSMN-VAD and CAM++ models, not an output of SenseVoiceSmall.
What licence is SenseVoice under?
The repository code is MIT. The SenseVoiceSmall weights on Hugging Face are published under the FunASR Model Open Source License Agreement, so read it before commercial use.
Can I use SenseVoice in the browser?
There are online demos on Hugging Face Spaces and ModelScope for trying it. For real use you run it yourself through FunASR.
Can WhipScribe transcribe Cantonese?
Yes. Cantonese is among the languages WhipScribe detects automatically, and the transcript stays in the language that was spoken. WhipScribe does not run SenseVoice, so compare a short clip if accuracy is critical.
What is SenseVoice?
SenseVoiceSmall transcribes Mandarin, Cantonese, English, Japanese and Korean, detects the language, and tags emotion and audio events. Its authors measure it at 15× the speed of Whisper-Large.
Is SenseVoice free?
Yes. SenseVoice is free and open source (free (MIT code; model under the FunASR model licence)).
How do I install SenseVoice?
With pip: `pip install -U "funasr>=1.3.26" # model: iic/SenseVoiceSmall`. You need Python and, for most audio tools, ffmpeg on your PATH first.
Is SenseVoice HIPAA compliant?
SenseVoice does not claim HIPAA compliance. For PHI, run it on infrastructure you control or choose a service that will sign a BAA.
What are the limitations of SenseVoice?
The released SenseVoiceSmall covers five languages only. Long recordings must be segmented (FSMN-VAD or its long_audio_no_vad.py script), and speaker diarization is a separate FunASR pipeline, not a model output.
Who is SenseVoice best for?
Developers transcribing Mandarin or Cantonese at volume, or who want emotion and audio-event tags alongside the text.
What should I know before choosing SenseVoice?
Part of the FunASR toolkit. Also inside apps like Whisper Notes and VoiceInk. For Cantonese recordings with no setup, a hosted service takes the file.
Does SenseVoice run offline?
SenseVoice runs on your own machine — it is open source, so nothing leaves the computer unless you configure it to.
Where is the SenseVoice source code?
On GitHub at github.com/QwenAudio/SenseVoice. It is maintained by FunAudioLLM (Alibaba Tongyi Lab).
Whipscribe is a hosted transcription service. If you want transcripts without running infrastructure, paste a URL or drop a file in the form below — you'll have a transcript in seconds.
Transcribe your recordings
Running SenseVoice yourself? Send the same file here and compare the transcript. Transcribe any language, any recording · Transcription services from $0.29 an hour · Transcribe from any meeting or recording platform · How to transcribe, step by step
Explore
All transcription tools · By language · By platform · By file format · How-to guides · Transcription services · Transcription API · Pricing
Have the recording already?
Skip the setup and send the file or the link: see transcribing a long MP3 or M4A, a podcast episode as text from its link and the SRT file generator.
Comparing tools before you pay?
If you would rather pay a service than run software yourself: see an honest roundup of the best transcription services, the cheapest way to transcribe an hour of audio and how our transcription services work.