How damvolkov/e-voice works
e-voice is a CPU-first Rust speech-to-text service that returns streaming or file transcripts with per-segment emotion. Its main workflow accepts audio through compatible HTTP/WebSocket APIs, normalizes audio, runs configurable speech-processing stages, and emits transcript events or a completed transcript. The README documents the pipeline and protocol capabilities; excerpts show gateway routing, batch decoding, and runner selection. Several internal workflow stages are documented but unsampled, so their detailed implementation links are limited to those documented relationships.
Main entry points: server.rs HTTP gateway · batch.rs File intake · live.rs Live intake · runner.rs Pipeline runner
Components 20
API gateway
- HTTP gateway
stt/src/api/server.rs - OpenAI APIs
stt/src/api/openai - Deepgram API
stt/src/api/deepgram/listen.rs - ElevenLabs Scribe
stt/src/api/elevenlabs/scribe.rs - Native stream
stt/src/api/native/stream.rs - File intake
stt/src/api/batch.rs - Live intake
stt/src/api/live.rs
Speech workflow
- Audio decoding
stt/src/core/audio.rs - Configurable stages
stt/src/workflow - Pipeline runner
stt/src/workflow/runner.rs - Segment sessions
stt/src/workflow/session.rs - Transcript events
stt/src/schema/transcript.rs
Runtime and configuration
- Request state
stt/src/api/state.rs - Pipeline config
stt/src/config/pipeline.rs - Service settings
stt/src/core/settings.rs - Startup lifecycle
stt/src/api/lifespan.rs
Audio client
- ecli commands
cli/src/stt.rs - Microphone input
cli/src/mic.rs - WAV playback
cli/src/wav.rs
Other components
- API clients
Connections 24
- API clients → HTTP gateway: sends audio
- HTTP gateway → OpenAI APIs: routes requests
stt/src/api/server.rs - HTTP gateway → Deepgram API: routes requests
stt/src/api/server.rs - HTTP gateway → ElevenLabs Scribe: routes requests
stt/src/api/server.rs - HTTP gateway → Native stream: routes requests
stt/src/api/server.rs - HTTP gateway → Startup lifecycle: starts service
stt/src/api/server.rs - OpenAI APIs → File intake: uses file intake
stt/src/api/openai/formats.rs - OpenAI APIs → Live intake: uses live intake
stt/src/api/openai/realtime.rs - Deepgram API → File intake: uses file intake
stt/src/api/deepgram/listen.rs - Deepgram API → Live intake: uses live intake
stt/src/api/deepgram/listen.rs - ElevenLabs Scribe → File intake: uses file intake
stt/src/api/elevenlabs/scribe.rs - Native stream → Live intake: uses live intake
stt/src/api/native/stream.rs - File intake → Audio decoding: decodes files
stt/src/api/batch.rs - File intake → Pipeline runner: transcribes audio
stt/src/api/batch.rs - Live intake → Pipeline runner: streams audio
stt/src/api/live.rs - Pipeline runner → Configurable stages: runs stages
- Configurable stages → Segment sessions: segments speech
stt/src/workflow/session.rs - Configurable stages → Transcript events: produces transcript
- Request state → Pipeline runner: holds runner
stt/src/api/state.rs - Service settings → Pipeline config: loads pipeline options
stt/src/core/settings.rs - Startup lifecycle → Request state: builds shared state
stt/src/api/state.rs - ecli commands → Microphone input: captures audio
cli/src/stt.rs - ecli commands → WAV playback: plays WAV input
cli/src/stt.rs - ecli commands → HTTP gateway: streams to gateway