dsh-voice-call

In AI agent interactions, agents often rely on passive text responses, lacking the ability to initiate communication. dsh-voice-call addresses this by giving DeepSeek Harness agents a "voice," enabling them to proactively make voice calls based on their judgment when deemed important, while humans retain control over answering, ensuring safety and respect.

Core capabilities include the offer_call tool for a call domain (ringing, accept/decline/postpone), local offline speech synthesis via CrispASR and Qwen3-TTS engines with 9 built-in voices (including Chinese dialects), and tools like speak and transcribe for voice playback and transcription. All audio actions require human initiation, prioritizing privacy.

This plugin is ideal for DeepSeek Harness users, especially developers, AI enthusiasts, or researchers, who want to explore agent voice interactions, build personalized assistants, test AI proactivity scenarios, or any environment requiring secure voice communication.