# QwenAudio/SenseVoice

Open-source SenseVoiceSmall model for Mandarin, Cantonese, English, Japanese, and Korean ASR, language ID, emotion recognition, and audio event detection.

Repository: https://github.com/QwenAudio/SenseVoice
Canonical: https://ross.abutalabs.com/products/sensevoice
Homepage: https://huggingface.co/spaces/FunAudioLLM/SenseVoice
Language: C
License: MIT
License Family: permissive
Topics: asr, speech-recognition, speech-to-text, cross-lingual, pytorch, speech-emotion-recognition, multilingual, audio-analysis, audio-event-detection, emotion-detection, voice-ai, whisper-alternative, funasr, language-identification, sensevoice, transcription, cantonese, llama-cpp, multilingual-asr, speech-understanding
Last push: 2026-08-18T02:13:34+00:00

## Health v2 (maintenance only)
Score: 88/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 98, release rhythm 94, longevity 56
- inputs: {"age_days": 791, "days_push": 16, "days_rel": 40, "gap_med": 6, "n_releases_24m": 6}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 9151, forks 812 (observed 2026-08-28T04:10:28.607821+00:00)

## What it is
SenseVoice is an open-source speech foundation model (SenseVoiceSmall) providing multilingual ASR, spoken language identification, speech emotion recognition, and audio event detection for Mandarin, Cantonese, English, Japanese, and Korean. It uses a non-autoregressive end-to-end framework for low-latency inference and integrates with FunASR for deployment and finetuning.

## Use cases
- transcribe speech to text in Mandarin, Cantonese, English, Japanese, or Korean
- detect emotion from speech audio
- identify the spoken language of an audio clip
- detect audio events like applause, laughter, or coughing
- find a fast low-latency alternative to Whisper for speech recognition
- finetune an ASR model on domain-specific audio samples

## When to choose
- you need fast, accurate multilingual ASR with extra emotion and audio-event tags
- your target languages are Mandarin, Cantonese, English, Japanese, or Korean
- you want low-latency non-autoregressive inference or easy finetuning via FunASR

## When to avoid
- you need speaker diarization as a single-model output (it requires composing separate FunASR VAD and CAM++ pipelines)
- you need ASR for languages beyond the five supported by the released checkpoint
- you need a tiny embedded deployment without GPU or C++/llama.cpp tooling

## Facets
- artifact type: library
- maturity: active
- function: speech-recognition, audio-processing, machine-learning, llm-inference
- domain: speech-processing, machine-learning
- platform: python, cpp, cross-platform
- tags: asr, speech-to-text, emotion-recognition, audio-event-detection, language-identification, multilingual, funasr, whisper-alternative, transcription, cantonese, audio, natural-language-processing, gpu

## Member repositories
- QwenAudio/SenseVoice (main) score 88

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:28.607821+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:23:24.024621+00:00, confidence not recorded.
  - readme: https://github.com/QwenAudio/SenseVoice (fetched 2026-08-28T04:10:28.607821+00:00, sha 86e30a2f1672)
  - homepage: https://huggingface.co/spaces/FunAudioLLM/SenseVoice (fetched 2026-08-29T08:23:27.860255+00:00, sha 00799e0fcae6)
- Data as of 2026-08-30T08:39:29.467469+00:00.