Summary

Add Deepgram Aura TTS as a first-class provider alongside OpenAI, ElevenLabs, and Edge TTS. Deepgram offers competitive pricing and low-latency speech synthesis via their Aura-2 voice models.

Motivation

• Deepgram is already used for STT (runtime.stt.transcribeAudioFile), so users may already have API keys
• Aura-2 voices are high quality at lower cost than OpenAI/ElevenLabs
• Deepgram natively supports PCM output ideal for telephony (voice-call plugin)
• Community requests for more TTS provider options

Scope

Full parity with existing providers across all surfaces:

• Config: messages.tts.deepgram block with apiKey, baseUrl, model
• Runtime: deepgramTTS() function with Token auth, auto-selection cascade (openai > elevenlabs > deepgram > edge)
• Output formats: MP3 (default), Opus (Telegram), linear16 PCM (telephony)
• Commands: /tts provider deepgram, /tts status shows Deepgram key status
• Gateway RPC: tts.status, tts.setProvider, tts.providers include Deepgram
• Model directives: [[tts:deepgram_model=...]] override support
• SecretRef: 3 credential paths for secrets configure/secrets apply
• Docs: TTS guide, config reference, voice-call plugin, plugin docs

API details

• Endpoint: POST {baseUrl}/v1/speak?model={model}&encoding={format}
• Auth: Authorization: Token {apiKey} (NOT Bearer)
• Body: {"text": "..."}
• Default model: aura-2-thalia-en
• Env fallback: DEEPGRAM_API_KEY