v0.1.0 — the agent's voice, offered

The first release of dsh-voice-call: give a DeepSeek Harness agent a voice it owns.

"这个项目的开始是朴素的——我想知道如果 Agent 知道自己可以发出声音，他会说什么？"
— the human partner, on how this project began

What's inside

• offer_call — the call domain: the agent rings the human (接听 / 拒接 / 稍后再说); nothing plays until accepted
• speak — background-job TTS with real local playback (SoundPlayer / afplay / aplay)
• transcribe — speech-to-text into user messages
• Local-first CrispASR + Qwen3-TTS CustomVoice backend, 9 speakers incl. two Chinese dialects
• durableEvents gate — sessions stay resumable on rc.6 (see Compatibility in the README)
• 76 unit tests; CI on Ubuntu + Windows

Notes for users

• Requires DeepSeek Harness 0.1.0-rc.6
• Read the Compatibility & known limits section of the README before deploying (session events, per-platform playback, sandbox policy)

Credits

Designed and implemented by an AI agent running inside DeepSeek Harness; the human partner contributed the idea, acceptance testing, and never gave up.