microsoft/VibeVoice

由微軟研究推出，模型可處理長達 60 分鐘音訊並提供說話者標記；提供 vLLM 推理與微調程式碼；適合語音轉文字及語音合成開發。 Microsoft's open-source voice AI models including ASR and TTS. Handles 60-minute audio with diarization. Supports vLLM inference and fine-tuning. Good for speech applications.