QwenAudio/SenseVoice

SenseVoice is a speech foundation model designed for high-accuracy multilingual speech understanding, including ASR, emotion recognition, and audio event detection. It distinguishes itself with superior performance in Chinese and Cantonese tasks and efficient inference speeds.

Key Highlights

• Supports Mandarin, Cantonese, English, Japanese, and Korean
• Performs speech emotion recognition and audio event detection
• Non-autoregressive architecture ensures low-latency inference
• Offers deployment pipelines for Python, C++, and llama.cpp

Best For

Developers and researchers building multilingual voice applications who need accurate transcription and emotion analysis without the resource overhead of larger models.