@houzi5940

Full-duplex voice for DeepSeek Harness (fork of haoku123/dsh-voice): streamed ASR, TTS with first-sentence/first-paragraph modes, barge-in, message replay, live settings.