Latest / Kana & Mari's SoundRepos (English)

sgl-project/sglang-omni
SGLang-Omni is a multi-stage serving runtime for multimodal, speech, text-to-speech, music-generation, and audio-transcription models. It coordinates heterogeneous pipeline stages, specialized schedulers, inter-stage tensor transport, model integrations, and OpenAI-compatible APIs, with support for streaming, voice cloning, diarization, routing, and comprehensive performance and accuracy benchmarks.
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.