How koljab/realtimestt works
RealtimeSTT is a speech-to-text library for microphone or application-fed audio, with optional live text, wake-word activation, and selectable recognition engines. Its main user path is to submit audio to a recorder, detect speech boundaries, produce realtime updates when enabled, and return a final transcript. The product also includes a packaged remote HTTP/WebSocket server and a browser streaming reference app. The graph follows documented capabilities and sampled module references; some internal call paths and server details are not shown in the excerpts, so those links are omitted rather than inferred.
Main entry points: audio_recorder.py Recorder API · recording.py Recording buffers · realtime.py Realtime transcription · transcription_engines Engine adapters
Components 23
Audio and activation
- Recorder API
RealtimeSTT/audio_recorder.py - Audio input
RealtimeSTT/audio_input.py - Voice activity
RealtimeSTT/core/voice_activity.py - Wake words
RealtimeSTT/core/wakeword.py - Recording buffers
RealtimeSTT/core/recording.py - Recorder state
RealtimeSTT/core/state.py
Transcription workflow
- Preview transcription
RealtimeSTT/core/preview_transcription.py - Realtime transcription
RealtimeSTT/core/realtime.py - Speech boundaries
RealtimeSTT/core/realtime_boundary_detector.py - Realtime text merge
RealtimeSTT/core/realtime_merge.py - Text stabilization
RealtimeSTT/core/realtime_text_stabilizer.py - Final transcription
RealtimeSTT/core/transcription.py - Tail finalization
RealtimeSTT/core/tail_transcription.py - Events and callbacks
RealtimeSTT/core/realtime_callbacks.py
Recognition engines
- Engine interfaces
RealtimeSTT/transcription_engines/base.py - Engine adapters
RealtimeSTT/transcription_engines - Model manifests
RealtimeSTT/model_manifests.py
Remote services
- Production HTTP/WebSocket API
RealtimeSTT_server/production_server.py - Browser streaming server
example_fastapi_server/server.py
Client integrations
- Recorder WebSocket client
RealtimeSTT/audio_recorder_client.py - Browser client
example_browserclient/client.js
Other components
- Application user
- ASR engine runtime
Connections 23
- Application user → Recorder API: configures and requests
- Application user → Audio input: feeds audio
- Recorder API → Voice activity: detects speech
- Recorder API → Wake words: optionally activates
RealtimeSTT/audio_recorder.py - Recorder API → Recording buffers: records utterance
RealtimeSTT/audio_recorder.py - Recording buffers → Preview transcription: requests live preview
RealtimeSTT/core/recording.py - Recording buffers → Realtime transcription: requests live updates
RealtimeSTT/core/recording.py - Realtime transcription → Speech boundaries: uses boundaries
- Realtime transcription → Realtime text merge: merges updates
- Realtime transcription → Text stabilization: stabilizes text
RealtimeSTT/core/initialization.py - Recording buffers → Final transcription: transcribes utterance
RealtimeSTT/core/recording.py - Final transcription → Tail finalization: uses speech tail
RealtimeSTT/core/recording.py - Recorder API → Events and callbacks: emits events
- Final transcription → Engine interfaces: uses engine interface
- Engine adapters → Engine interfaces: implements interface
RealtimeSTT/transcription_engines/funasr_engine.py - Engine adapters → ASR engine runtime: runs recognition
- Engine adapters → Model manifests: resolves model profiles
RealtimeSTT/transcription_engines/nemotron_engine.py - Recorder WebSocket client → Audio input: captures microphone audio
RealtimeSTT/audio_recorder_client.py - Recorder WebSocket client → Production HTTP/WebSocket API: streams to server
- Application user → Production HTTP/WebSocket API: uses remote API
- Browser client → Browser streaming server: streams audio
- Browser streaming server → Recorder API: uses recorder
example_fastapi_server/server.py - Production HTTP/WebSocket API → Engine adapters: provides transcription