Chunk offline audio decoding to prevent ONNX SIGTRAP crashes

[AmethystLiang]

Summary

This PR resolves SIGTRAP out-of-memory crashes that occur during long offline dictation/speech-to-text (STT) sessions.

Previously, the offline recognizer buffered all audio to decode in one single shot. For long recordings, ONNX Runtime's arena allocations scaled with buffer length, eventually hitting Chromium's 2 GiB allocation limit (#7925) and crashing the entire application.

This change introduces OfflineAudioChunker to break the audio stream into 30-second chunks before decoding. To prevent transcription degradation at boundary splits, it searches for quiet pauses (inter-word intervals) near the boundary.

Screenshots

No visual change.

Testing

• pnpm lint
• pnpm typecheck
• pnpm test
• pnpm build
• Added or updated high-quality tests that would catch regressions, or explained why tests were not needed

Added comprehensive unit tests in stt-offline-audio-chunker.test.ts covering buffering, boundary limits, silence detection splits, value conservation, and empty flush behavior.

AI Review Report

• Main risks checked: ONNX memory consumption, buffer sizing/overflows, audio quality degradation at split boundaries, and stream recreation logic.
• Worker/Stream Recreation: Verified that we recreate the offline stream after each decoded chunk (sherpa.createOfflineStream(recognizer)) to keep the recognizer's internal state isolated and correct.
• Cross-Platform Compatibility: Explicitly checked cross-platform compatibility across macOS, Linux, and Windows. The chunking logic is implemented purely in platform-agnostic TypeScript/Float32Array operations and is executed inside Node.js STT worker threads, ensuring consistent behavior across all supported Electron platforms.

Security Audit

• Input Handling: Audio streams are parsed and chunked within the isolated Node.js worker thread.
• IPC & Execution: No external commands, subprocesses, path manipulation, or IPC boundaries are touched.

Notes

The 2 GiB Chromium allocation limit is a platform-level constraint affecting all Chromium/Electron environments. This worker-level chunking bypasses the restriction uniformly without requiring complex environment configurations.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 99fb757e-3efe-4349-a9e7-4f7639432dc3

📥 Commits

Reviewing files that changed from the base of the PR and between ee823b7 and c2ec7ba.

📒 Files selected for processing (3)

• src/main/speech/stt-offline-audio-chunker.test.ts
• src/main/speech/stt-offline-audio-chunker.ts
• src/main/speech/stt-worker.ts

---

📝 Walkthrough

Walkthrough

This PR adds a new OfflineAudioChunker utility (src/main/speech/stt-offline-audio-chunker.ts) that buffers incoming audio and emits bounded chunks (capped at OFFLINE_DECODE_CHUNK_SECONDS, 30s), selecting split points at low-energy boundaries to avoid cutting mid-speech, with flush() to drain remaining buffered audio. A corresponding Vitest test suite validates buffering, chunk-size limits, splitting, and sample preservation. stt-worker.ts is updated to replace its prior full-buffer offline decoding approach with this chunker: initializing it during init, pushing samples and decoding ready chunks incrementally during feed, flushing and decoding remaining audio during stop, and clearing the chunker during teardown, using a new decodeOfflineChunk helper that recreates the Sherpa offline stream after each decode.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and accurately summarizes the main change: chunking offline audio decoding to avoid SIGTRAP crashes.
Description check	✅ Passed	The description matches the template well and includes Summary, Screenshots, Testing, AI Review Report, Security Audit, and Notes.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

Warning

Review ran into problems

🔥 Problems

Git: Failed to clone repository. Please run the @coderabbitai full review command to re-trigger a full review. If the issue persists, set path_filters to include or exclude specific files.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}