Full-duplex voice plugin for DeepSeek Harness
qishuilalala/dsh-voice-mode · skills
DeepSeek Harness full-duplex voice plug-in - real-time recognition · Read aloud by sentence · Interrupt as soon as you open your mouth
★ 10GitHub stars
2Forks
2026-09-07Last updated
JavaScriptLanguage
MITLicense
Key features
- Read aloudDefault Edge cloud
- You say - the words are produced in real time while speaking (streaming recognition), and the words are automatically sent after a pause of about 700ms
- Settings → Plugins → Plug-in Configuration → Voice-mode.
- Local VITS (Pure Chinese 5 speakers)/Kokoro (Chinese-English mixed reading 103 sounds, int8 default/fp32 optional), independent child process, crash self-healing
- Streaming recognitionzipformer2 streaming (writing while speaking) + SenseVoice finalization (with punctuation/number normalization)
Install command
dsh plugin --profile web add github:qishuilalala/dsh-voice-mode