QwenAudio /SenseVoice
Open-source SenseVoiceSmall model for Mandarin, Cantonese, English, Japanese, and Korean ASR, language ID, emotion recognition, and audio event detection.
★ 9,758⑂ 849👁 63⑃ 84● 266▤ 7
+5
Daily
9,758
stars
849
forks
63
watchers
84
prs
266
issues
7
releases
README
(简体中文|English|日本語) Introduction SenseVoice is a speech foundation model with multiple speech understanding capabilities, including automatic speech recognition (ASR), spoken language identification (LID), speech emotion recognition (SER), and audio event detection (AED). [//]: ( ) Homepage | What's News | Benchmarks | Install | Usage | Community Model Zoo: modelscope, huggingface, paper Online Demo: modelscope demo, huggingface space Released checkpoint scope: SenseVoiceSmall supports ASR and language ID for Mandarin, Cantonese, English, Japanese, and Korean, together with emotion and audio-event tags. Speaker diarization is a composed FunASR pipeline using separate FSMN-VAD and CAM++ models; it is not an output of the SenseVoiceSmall checkpoint itself. Highlights 🎯 SenseVoice focuses on high-accuracy multilingual speech recognition, speech emotion recognition, and audio event detection. - Research scope vs released checkpoint: The broader SenseVoice work reports training on more than 400,000 hours and support for more than 50 languages. The released SenseVoiceSmall checkpoint linked above supports Mandarin, Cantonese, English, Japanese, and Korean; the benchmark comparisons below are task- and language-specific. - Rich transcribe: - Possess excellent emotion recognition capabilities, achieving and surpassing the effectiveness of the current best emotion recognition models on test data. - Offer sound event detection capabilities, supporting the detection of various common human-computer interaction events such as bgm, applause, laughter, crying, coughing, and sneezing. - Efficient Inference: SenseVoiceSmall uses a non-autoregressive end-to-end framework for low-latency inference; see the benchmark setup and comparison below. - Convenient Finetuning: Provide convenient finetuning scripts and strategies, allowing users to easily address long-tail sample issues according to their business scenarios. - Service Deployment: Offer service deployment pipeline, supporting multi-concurrent requests, with client-side languages including Python, C++, HTML, Java, and C, among others. What's New 🔥 - Current deployment path: install funasr==1.4.14 for SenseVoice Python, OpenAI-compatible service, and container workflows. Release notes · SenseVoice releases - Long audio without VAD: long_audio_no_vad.py uses bounded overlapping windows and preserves raw chunk outputs, so hour-scale recordings do not require one unbounded GPU allocation. Run it - - Integrated diarization alternative: the wider FunASR ecosystem supports OpenMOSS/MOSS-Transcribe-Diarize for offline transcription, timestamps, and anonymous speaker labels without composing external VAD and speaker models. Deployment guide - See Releases for the complete version history. Benchmarks 📝 Multilingual Speech Recognition We compared the performance of multilingual speech recognition between SenseVoice and Whisper on open-source benchmark datasets, includ
Top issues
#339[Bug]: Unable to pull Docker image or start Docker service9 comments · 2026-08-07#215时间戳数量和字符数量不匹配【词粒度模式返回时间戳】6 comments · 2025-04-24#212CASIA 数据集 SER 性能不能复现,测试结果只有34,而不是report的706 comments · 2025-04-15#7torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 214.66 GiB.6 comments · 2024-07-05#343在edge device 中,只能使用cpu推理吗5 comments · 2026-08-18#291使用VAD + CAM++ speaker model + punctuation model 返回结果是空的5 comments · 2026-06-10#360Text doesn't match any voice clip3 comments · 2026-10-07#334Clarification: does MODEL_LICENSE §3 "reference and learning purposes only" restrict commercial use? / 关于 §3 与商用边界的确认3 comments · 2026-07-26
Discussions
#345altgo:在桌面端离线跑 SenseVoice-Small int8 做语音输入1 comments · 2026-08-24#336SenseVoice lands in Orca v1.4.159 RC: offline dictation on macOS, Linux, and Windows0 comments · 2026-07-27#335SenseVoice deployment hub: llama.cpp, ONNX CPU, and private APIs0 comments · 2026-07-26#333runtime-llamacpp-v0.1.9: complete packages and safe source snapshot0 comments · 2026-07-24#318MyVoiceTyping: a local-first Chinese voice typing app powered by SenseVoice/FunASR1 comments · 2026-07-16
Latest release
FunASR llama.cpp runtime v0.2.1 runtime-llamacpp-v0.2.1 2026-08-27Prebuilt self-contained binaries for the FunASR llama.cpp / GGUF runtime, including SenseVoice, Fun-ASR-Nano, and Paraformer with built-in FSMN-VAD. v0.2.1 - Fixes Vulkan selection when a requested device is reported as an integrated GPU (GGML_BACKEND_DEVICE_TYPE_IGPU). - Prefers a matching discrete GPU and falls back to a matching integrated GPU. - Provides nine desktop archives for Linux, macOS, and Windows, including Vulkan and Windows CUDA variants. - The Windows Vulkan package is ready for Radeon 780M retesting. This selector fix does not claim to resolve the separate Radeon RX 9070 XT