Transcript text not appearing for any recordings

[hyper-charlie]

Problem

User reports that transcripts are completely blank across all recordings. Audio playback works fine under the Transcript tab, but no transcript text is displayed.

Steps to Reproduce

1. Create a new recording (40+ minutes tested)
2. Allow recording to complete
3. Navigate to the Transcript tab
4. Audio plays back normally, but transcript area is empty

Expected Behavior

Transcript text should appear in the transcript view after recording completes.

Actual Behavior

Transcript area remains completely blank for every recording. Audio is present and playable.

Environment

• App Version: 1.0.11
• Build: 0b5cb5841b9dd0a28f4e4c4585a42cb3ea86828959
• Platform: macOS (aarch64)
• OS Version: macOS 26.3
• Locale: it-IT
• User ID: 9cdefca2-9bbf-4816-89fe-abe02b921b63

Additional Notes

• Issue affects all recordings, not just one
• This is a consistent, reproducible problem
• Audio capture is working correctly

[devin-ai-integration]

Codebase Exploration: Transcript Display Issue

Transcript Data Flow Summary

Live recording path:

1. useStartListening (src/stt/useStartListening.ts) creates a transcript row in TinyBase with words: "[]"
2. Listener plugin captures audio → STT provider produces StreamResponse
3. Backend emits SessionDataEvent::StreamResponse (crates/listener-core/src/events.rs)
4. Frontend handleDataEvent (store/zustand/listener/general.ts:317) calls handleTranscriptResponse
5. handleTranscriptResponse (store/zustand/listener/transcript.ts:157) filters for response.type === "Results", transforms words, and calls handlePersist callback to write to TinyBase

Viewing saved recordings path:

1. loadAllSessionData / loadSingleSession (store/tinybase/persister/session/load/index.ts) scans for transcript.json files
2. processTranscriptFile (store/tinybase/persister/session/load/transcript.ts) parses JSON and populates TinyBase
3. useFinalWords hook (session/components/note-input/transcript/shared/hooks.ts:21) reads words cell from TinyBase
4. buildSegments → RenderTranscript renders the segments

Key Files

File	Role
apps/desktop/src/stt/useStartListening.ts	Wires persist callback, creates transcript row
apps/desktop/src/stt/useSTTConnection.ts	Determines STT provider/model/connection
apps/desktop/src/store/zustand/listener/transcript.ts	Processes incoming StreamResponse into words
apps/desktop/src/store/zustand/listener/general.ts	Session lifecycle, event handling
apps/desktop/src/session/components/note-input/transcript/shared/index.tsx	TranscriptContainer — main transcript display
apps/desktop/src/session/components/note-input/transcript/shared/render-transcript.tsx	RenderTranscript — renders word segments
apps/desktop/src/store/tinybase/persister/session/load/transcript.ts	Loads transcript from disk into TinyBase
apps/desktop/src/store/tinybase/persister/session/save/transcript.ts	Saves transcript from TinyBase to disk
crates/listener-core/src/actors/listener/mod.rs	Handles StreamResponse from STT, emits to frontend
crates/owhisper-interface/src/stream.rs	StreamResponse enum definition

Potential Root Causes

1. STT Provider Not Returning Results (affects new recordings only)

handleTranscriptResponse (transcript.ts:158) silently returns if response.type !== "Results". If the STT provider returns errors or non-result messages, no words accumulate. The user's locale is it-IT — if their spoken_languages setting doesn't align with what their STT provider supports, the provider may not return any transcript results.

Key question: What STT provider is this user configured with? (Hyprnote cloud, local Whisper, Deepgram, etc.)

2. STT Connection Null — Recording Starts But No STT

In useSTTConnection.ts, the connection returns null in several cases:

• No current_stt_provider or current_stt_model configured
• Cloud model: no active auth session or not pro subscriber
• Local model: not downloaded or server not ready
• Third-party: missing baseUrl or apiKey

If conn is null, useStartListening (line 36-38) returns early without starting the session. However, this would likely also prevent audio capture, which contradicts the user report that audio plays back fine.

3. Recent v1.0.11 Changes — Possible Regressions

Commits in the v1.0.11 build (0b5cb5841):

• 0b62cd920 — "Fix proxy forwarding Hyprnote control messages to upstream providers" (Fix proxy forwarding Hyprnote control messages to upstream providers #4364): Changes to transcribe-proxy affect how control messages (KeepAlive, Finalize, CloseStream) are forwarded to upstream STT providers. If the user is on the Hyprnote cloud proxy, a regression here could cause the upstream provider to not produce transcript results. This is server-side, so the fix/regression would depend on the deployed API version.

• 47aa41830 / 2f314f847 — Resampler refactor + rubato v1 bump: The resampler was extracted into a new crate and rubato was bumped to v1. If there's an audio resampling regression, the STT provider would receive malformed/silent audio and return no transcription. This could explain blank transcripts for all new recordings.

4. Known Display Gap — words=[] Case

There's a TODO at TranscriptContainer line 140:

// TOOD: this can't handle words=[]

If transcript rows exist in TinyBase but have empty words, transcriptIds.length > 0 passes the guard at line 141, so the empty state isn't shown. Instead, each RenderTranscript returns null (render-transcript.tsx:87-89) because segments.length === 0. This results in a blank area with no feedback to the user.

This is relevant if words are somehow not being persisted (e.g., handlePersist not wired, or the persist callback silently failing).

5. Disk Persistence Failure (affects saved recordings)

If the save pipeline (buildTranscriptSaveOps) fails silently, transcript rows would have words: "[]" on disk even though they had words in memory during recording. On next app launch, processTranscriptFile loads the empty words. The user would see transcripts during recording (via partialWords) but blank transcripts after reopening the session.

Diagnosis Approach

Critical question: Does the issue affect only completed/saved recordings, or also recordings in progress?

• If in-progress recordings also show no text → STT pipeline issue (causes 1, 2, or 3)
• If text appears during recording but disappears after → persistence issue (cause 5)
• If all recordings including previously-working ones are blank → loading/display issue (cause 4)

Useful information to request from the user:

1. STT provider and model being used (Settings → Transcription)
2. Whether any text flashes during active recording before disappearing
3. Whether this started specifically after updating to v1.0.11
4. Console/log output (if accessible) — errors from [listener] or [SessionPersister] prefixes would be very telling
5. Whether the transcript.json files in the user's data directory contain words arrays with data or empty arrays