@harness-flow/dsh-voice

配置

- id: dsh-voice
  config:
    asr:
      engine: auto          # auto | qwen | mimo | local
      language: zh
      # qwen: { apiKeyEnv: DASHSCOPE_API_KEY, model: qwen3-asr-flash, baseUrl: https://dashscope.aliyuncs.com/compatible-mode/v1 }
      # mimo: { apiKeyEnv: MIMO_API_KEY, model: MiMo-V2.5-ASR, baseUrl: https://api.mimo.mi.com/v1 }
      # local: { kind: sherpa-onnx | whisper-bin, modelType: paraformer, modelDir: ~/.dsh/voice-models/paraformer-zh-small, model: model.int8.onnx, bin: ... }
    tts:
      engine: auto          # auto | qwen | mimo | local
      voice: Cherry
      # qwen: { apiKeyEnv: DASHSCOPE_API_KEY, model: qwen3-tts-flash, baseUrl: https://dashscope.aliyuncs.com/api/v1 }
      # mimo: { apiKeyEnv: MIMO_API_KEY, model: MiMo-V2.5-TTS, baseUrl: https://api.mimo.mi.com/v1 }
      # local: { kind: sherpa-onnx | piper-bin | say, modelDir: ..., bin: ..., model: ... }
    audioDir: ~/.dsh/voice
    listenTimeoutSec: 120

• 云引擎密钥只在调用时从引用的环境变量读取，配置里只写变量名，永不写密钥。
• sherpa-onnx 需要 profile 里安装 sherpa-onnx-node 并配置模型目录；whisper-bin/piper-bin 指向 whisper.cpp / piper 的可执行文件与模型。
• 本地 MiMo（MiMo-Audio-7B 经 vLLM-Omni 等 OpenAI 兼容网关）通过 tts.mimo.baseUrl / asr.mimo.baseUrl 指向本机地址即可复用同一实现。