ASR and TTS Voice Configuration Assistant

Problem

After connecting an LLM chat bot to Feishu, Telegram, WeCom, and other channels, voice messages can stall the workflow: audio attachments are not transcribed, and text replies lack natural spoken output. This skill turns TTS and ASR setup into checkable, selectable, installable steps instead of falling back to “I cannot process voice.”

How It Works

It routes user intent to TTS voice replies or ASR speech-to-text:

• TTS: checks ready, partial, or not_configured status, then presents edge-tts-universal or Tencent Cloud TTS, and requires a Gateway restart after installation.
• ASR: treats audio attachments or explicit transcription requests as triggers, probes machine resources, and selects local Whisper or Tencent Cloud ASR with model-size guidance.

The workflow presents cost, privacy, latency, and deployment trade-offs first, then executes only after explicit selection.

Boundaries and Notes

This is a configuration and deployment helper, not a speech model itself. Voice capabilities still depend on underlying plugins, credentials, service restarts, and vendor quotas. The free local path suits offline or low-cost setups, while cloud options require keys. ASR should not trigger when JSON appears only in documentation, code samples, or quoted content.