[Feature]: optional TTS replies for voice/audio prompts

[shanekunz]

Problem

The bot already supports voice/audio input via STT, which is great for mobile use.
A natural follow-up for that workflow is optional TTS output: when a user sends a Telegram voice/audio message toggle TTS on with /tts, the bot could return the normal text reply plus an audio rendering of that same final assistant response.
This would improve hands-free/mobile usability without changing the normal text-first workflow. (idea from OpenClaw).

Proposal

• Keep normal text prompts exactly as they are today: text reply only
• For Telegram voice / audio input:
  • transcribe with the existing STT flow
  • send the normal final text response
  • optionally send a TTS audio file of that exact final assistant text
• Make it opt-in via env config make a /tts toggle, disabled by default

Proposed config

Something like:
- TTS_ENABLED=false use /tts toggle instead

• TTS_API_URL= (fallback to STT_API_URL if unset)
• TTS_API_KEY= (fallback to STT_API_KEY if unset)
• TTS_MODEL=gpt-4o-mini-tts
• TTS_VOICE=alloy

Scope / guardrails

To keep this small and low-risk:

• no change for text-origin prompts
• no streaming spoken output
• just one final audio file after the normal text reply

Why this seems aligned

This stays within the current single-chat / predictable interaction model in CONCEPT.md:

• it does not add parallelism or group-specific behavior
• it only extends the existing voice-input path
• it remains optional and disabled by default

Implementation notes

I already prototyped this locally my fork and it was pretty contained:

• small TTS client modeled after the existing STT client
• lightweight tracking so only audio-origin prompts trigger TTS
• hook into the final assistant completion path after the normal text reply
• docs + tests included

Done criteria (optional)

• When sending a voice memo/file, and .env has it enabled and enabled via /tts, bot responds with text output and then an audio file of that text
  - When sending a text input, bot always replies as before with text regardless of configuration.

[grinev]

@shanekunz thats great idea. But I think audio replies should be switched with command or reply keyboard. Because sometimes you need this mode, sometimes not and I you need quickly switch it without restarting server. For now it's easier to implement command which turns on voice mode. What do yo think about it?

[shanekunz]

@grinev Yes, I like that idea as well. I can make it a slash command.