Scale Cloud Whisper timeout for long recordings

[katanumahotori]

User description

Summary

• replace the fixed 15s coordinator timeout for Cloud Whisper-compatible dictation with an audio-duration-aware timeout
• apply the same timeout to the existing history retranscription path using the archived PCM duration
• add timeout boundary tests for short, zero-length, long, capped, and exact-cap recordings

Why

Long Cloud Whisper-compatible recordings can take longer than the existing 15s coordinator guard to transcribe. In that case the recording is marked as a timeout even though the provider may still be working. This is especially visible for multi-minute recordings.

The new timeout is:

• floor: 30s
• scaled: ceil(audio_secs * 0.5) + 15
• cap: 300s

For example, a 422s recording now gets a 226s timeout instead of 15s.

The same timeout is also used when retranscribing an archived recording, where the exact PCM duration is available.

Validation

• cargo test cloud_whisper_timeout --lib passed: 5 tests
• cargo test passed: 521 tests
• cargo build --release passed
• Claude review: no blockers, ready for PR
• Codex review: no blockers, ready for PR

Note: cargo fmt --check currently reports unrelated existing formatting diffs in upstream files outside this PR, so I did not apply broad formatting changes.

---

PR Type

Enhancement, Bug fix

---

Description

• Replace fixed 15s coordinator timeout with audio-duration-aware timeout for Cloud Whisper

• Apply same timeout to retranscription path and add Apple Speech retranscription support

• Add boundary tests for short, zero, long, capped, and exact-cap recordings

---

Diagram Walkthrough

flowchart LR
  AudioLength["Audio length (seconds)"]
  ComputeTime["Compute timeout: ceil(len*0.5)+15, clamped [30,300]"]
  Transcribe["Cloud Whisper transcribe with timeout"]
  Retranscribe["Retranscribe archived PCM with same timeout"]
  AudioLength --> ComputeTime
  ComputeTime --> Transcribe
  ComputeTime --> Retranscribe

Loading

File Walkthrough

	Relevant files
Enhancement	coordinator.rs Use dynamic timeout in retranscribe_pcm and add Apple Speech openless-all/app/src-tauri/src/coordinator.rs Added import of pcm module to compute audio duration Modified retranscribe_pcm to use cloud_whisper_transcribe_timeout for Whisper provider Added Apple Speech retranscription timeout using local_qwen_transcribe_timeout +18/-6    asr_setup.rs Add timeout constants and scaling function for Cloud Whisper openless-all/app/src-tauri/src/coordinator/asr_setup.rs Defined CLOUD_WHISPER_MIN_TIMEOUT_SECS (30) and CLOUD_WHISPER_MAX_TIMEOUT_SECS (300) Implemented cloud_whisper_transcribe_timeout function with formula ceil(audio_secs * 0.5) + 15, clamped between min and max +13/-0
Bug fix	dictation_end.rs Use audio-aware timeout in Cloud Whisper end_session          openless-all/app/src-tauri/src/coordinator/dictation_end.rs Replaced fixed COORDINATOR_GLOBAL_TIMEOUT_SECS with dynamic cloud_whisper_transcribe_timeout in end_session for Whisper provider Added logging of audio duration and computed timeout +8/-3
Tests	tests.rs Add unit tests for cloud_whisper_transcribe_timeout            openless-all/app/src-tauri/src/coordinator/tests.rs Added 5 test cases: floor for short recordings, zero duration, scaling for 422s, cap at 1000s, exact cap boundary at 570s +42/-0

---

[github-actions]

PR Reviewer Guide 🔍

(Review updated until commit c7a65b9)

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🧪 PR contains tests

🔒 No security concerns identified

⚡ No major issues detected

[katanumahotori]

Clarification: this PR is not intended to implement the full history recording/retranscription feature request. That functionality is already present on beta. This PR only fixes the remaining fixed 15s Cloud Whisper timeout in normal dictation and in the existing retranscription path.

[github-actions]

Persistent review updated to latest commit c7a65b9