Live transcription: idle onboarding (no highlight) + real-time speech#18
Live transcription: idle onboarding (no highlight) + real-time speech#18
Conversation
… recording\n\n- Replace hard-coded demo transcript with onboarding instruction\n- Chip + blinking cursor now appear only when recording (isCursorVisible)\n- Toggle cursor visibility on recording lifecycle transitions\n- No changes to services; speech stream continues to update transcript in real time
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting
| return ChangeNotifierProvider<VoiceToTextModelState>( | ||
| create: (_) => VoiceToTextModelState( | ||
| create: (context) => VoiceToTextModelState( | ||
| initialTranscript: const [ | ||
| 'Herman', | ||
| 'is', | ||
| 'just', | ||
| 'like', | ||
| 'Tap', | ||
| 'the', | ||
| 'rest', | ||
| 'microphone', | ||
| 'and', | ||
| 'tell', | ||
| 'your', | ||
| 'AI', | ||
| 'Chief', | ||
| 'of', | ||
| 'us.', | ||
| 'Everyday', | ||
| 'he', | ||
| 'has', | ||
| 'Staff', | ||
| 'what', | ||
| 'to', | ||
| 'make', | ||
| 'all', | ||
| 'kin', | ||
| 'do', | ||
| '—', | ||
| 'send', | ||
| 'a', | ||
| 'Slack', | ||
| 'message,', | ||
| 'draft', | ||
| 'an', | ||
| 'email,', | ||
| 'check', | ||
| 'the', | ||
| 'calendar,', | ||
| 'or', | ||
| 'follow', | ||
| 'up', | ||
| 'for', | ||
| 'you.', | ||
| ], | ||
| initialActiveWordIndex: 14, | ||
| audio: context.read<AudioCaptureService>(), | ||
| speech: context.read<SpeechRecognitionService>(), |
There was a problem hiding this comment.
[P1] Start idle screen with cursor still visible
The voice screen builds VoiceToTextModelState without overriding initialCursorVisible. Because the model’s default is true, the onboarding text still renders the active-word chip the moment the screen appears even though the recording state is idle. The headline feature here is that the chip should show only while recording, so the model needs to start with initialCursorVisible: false (or change the default) before displaying the onboarding copy.
Useful? React with 👍 / 👎.
| @override | ||
| void dispose() { | ||
| _timer?.cancel(); | ||
| _waveformController.close(); | ||
| _teardownStreams(); |
There was a problem hiding this comment.
[P1] Disposing model leaves audio and speech services running
VoiceToTextModelState.dispose currently cancels timers and subscriptions but never stops the injected AudioCaptureService or SpeechRecognitionService. If the user navigates away while recording, the microphone and speech recognizer continue running in the background, because only _teardownStreams() is invoked. Disposing the model should call audio.stop() and speech.cancel() (or a similar teardown) to release system resources and avoid capturing audio after the UI disappears.
Useful? React with 👍 / 👎.
…n\n- initialCursorVisible=false and active index reset for idle\n- On startRecording, clear transcript and enable cursor so incoming speech streams immediately\n- Keep highlight strictly tied to recording state
fbraza
changed the title
…ediately on record\n\n- Always initialize speech engine if not initialized before listening\n- Ignore empty partials; stream recognizedWords when available\n- Clear onboarding transcript at startRecording and enable cursor\n- Startup state uses no cursor/active highlighting
…permission\n\n- Resolve missing live text by ensuring engine is initialized and localeId is set\n- Configure listenFor/pauseFor to keep the session open\n- Add android INTERNET permission for cloud-backed recognition
1 participant
Summary\nThis PR ensures live transcription text displays in real time during recording and no text is highlighted when idle.\n\n### Fixes\n- Clear onboarding transcript on startRecording so incoming speech is shown immediately.\n- Cursor/active chip appears only while recording; hidden when idle/paused/stopped.\n- Speech recognition now always initializes if needed before listening, and ignores empty partials.\n\n### Files\n- lib/core/services/speech_recognition.dart — init guard, partial handling.\n- lib/features/voice_to_text/view/voice_to_text_model.dart — clear transcript and toggle cursor on start, keep highlight behavior tied to lifecycle.\n- lib/features/voice_to_text/view/voice_to_text_screen.dart — startup state with no highlight.\n\n### Verification\n- flutter analyze (clean), flutter test (green).\n- On device/emulator: press mic, speak; text area streams recognized words; pausing/stopping hides the cursor and keeps text.\n