🎧 Multimodal ingest covers speech, speaker identity, sound events, multi-speaker conversations and music, bundling ASR, voiceprint, scene and emotion detection
🤖 Model family with #Qwen25Omni, #Qwen3Omni and #StepAudio2Mini, trained on the open ChatMem-400K dataset through a three-stage OPD pipeline, finetuning code included