Adds Smallest AI as a new TTS and STT provider for Clawdbot:
TTS (Text-to-Speech):
- Lightning v3.1 model for fast, low-latency speech synthesis
- Support for waves model (higher quality)
- Native mulaw @ 8kHz output for telephony (no resampling needed)
- 30+ voice options (lauren, emily, jasmine, arman, etc.)
- Configurable speed, consistency, similarity, enhancement
STT (Speech-to-Text):
- Pulse model for accurate audio transcription
- Automatic language detection with optional language hint
- Raw audio bytes API for efficient streaming
Configuration:
- New provider: messages.tts.provider = 'smallestai'
- Environment variable: SMALLEST_API_KEY
- Full config options in messages.tts.smallestai
Tested with voice-to-voice flow on Telegram.
AI-assisted (Claude). Fully tested locally.
