digiteinfotech / kairon / 35689131303

92%

Build Type

push

github

Committed by web-flow

Commit Message

Voice integration exotel (#2436)

* Add Exotel voice scaffolding (P0): provider-agnostic STT/TTS layer + config

Scaffolding only — provider-agnostic, no vendor SDKs, no call behavior yet.
Exotel is added alongside Twilio (not a replacement); everything is bot-scoped.

- constants: VoiceProviderTypes.exotel + VoiceServiceType/STTProviderTypes/TTSProviderTypes
- metadata/integrations.yml: voice_channels.exotel; metadata-driven secret_fields
  (validate_voice_provider now reads secret_fields from metadata, Twilio-compatible)
- metadata/stt_providers.yml, tts_providers.yml: versioned provider registries
  (models per sample-rate, languages, required/secret fields)
- system.yaml voice: global section (8kHz default, 320-byte chunk rule,
  fallback order, per-provider global keys)
- VoiceProviderCredential: bot-scoped, encrypted BYO-provider credentials
- BotSettings.voice (VoiceIntegrationSettings): per-bot STT/TTS selection
- BaseSTT/BaseTTS ABCs + STTFactory/TTSFactory registries
- Unit tests: metadata load, Exotel channel validation, factories, settings

* Add Exotel streaming voice call loop (P1): WS gateway + Sarvam STT + Polly TTS

- exotel/protocol.py: Exotel Voice Streaming frame parse/build (connected/start/
  media/dtmf/mark/stop/clear), base64<->PCM helpers
- exotel/audio.py: AudioChunker re-framing TTS output to Exotel's 320-byte-multiple
  rule with silence padding; RMS/silence detection
- exotel/session.py: ExotelCallSession orchestrating caller-audio->STT->agent->TTS
  with real-time pacing, welcome greeting, and hangup handling
- stt/sarvam.py, tts/polly.py: concrete streaming adapters (vendor SDKs lazy-imported)
- stt/tts factories: lazy self-registration of built-in adapters
- chat WebSocket route + ExotelStreamHandler bridging to AgentProcessor/VoiceOutput
- unit tests for protocol, audio, Sarvam parsing, and the full session loop
- add boto3/websockets to prod requirements

* Add voice resilience & quality layer (P2): fallback, TTS cach... (continued)

Coverage Stats

1287 of 1349 new or added lines in 29 files covered. (95.4%)

12 existing lines in 1 file now uncovered.

34175 of 37251 relevant lines covered (91.74%)

0.92 hits per line