fix(asr): Whisper/OpenRouter 动态超时 + 全局兜底 15→30s

[H-Chris233]

User description

问题

Closes #721

OpenRouter Whisper 长语音被 15s 全局超时硬截断。

方案

双重保险：

1. 全局常量 15→30s：`COORDINATOR_GLOBAL_TIMEOUT_SECS` 提升为云端 ASR 留网络预算
2. Whisper 动态超时：`whisper_transcribe_timeout(audio_secs)` = `max(30, ceil(audio×0.5)+20)`，参考本地 Qwen 的 `local_qwen_transcribe_timeout`

音频	旧超时	新超时
10s	15s	30s
30s	15s ❌	35s ✅
60s	15s ❌	50s ✅
90s	15s ❌	65s ✅

改动

3 文件，+32 / −7 行，`cargo check` 通过。

---

PR Type

Bug fix, Enhancement

---

Description

• Increase global ASR timeout from 15s to 30s

• Add dynamic timeout for Whisper/OpenRouter based on audio duration

• Log audio duration and timeout for debugging

• Add unit tests covering new timeout logic

---

Diagram Walkthrough

flowchart LR
  endSession["End Session (Whisper)"] -- audio_secs from buffer_duration_ms() --> calc["whisper_transcribe_timeout(audio_secs)"]
  calc -- "timeout_duration = max(30, ceil(audio*0.5)+20)" --> call["tokio::time::timeout(timeout_duration, w.transcribe())"]
  call -- "Ok(r)" --> transcript["Return transcript"]
  call -- "Err(_)" --> log["Log timeout with audio_secs"]

Loading

File Walkthrough

	Relevant files
Enhancement	whisper.rs Expose buffer duration for timeout calculation                      openless-all/app/src-tauri/src/asr/whisper.rs Add buffer_duration_ms() method to WhisperBatchASR to expose audio duration Allows coordinator to compute dynamic timeout before transcribing +6/-0      coordinator.rs Implement Whisper timeout function and update tests            openless-all/app/src-tauri/src/coordinator.rs Raise COORDINATOR_GLOBAL_TIMEOUT_SECS from 15 to 30 Add whisper_transcribe_timeout() function with formula max(30, ceil(audio*0.5)+20) Update local_qwen_timeout tests to align with new global timeout Add 3 new unit tests for Whisper timeout (short audio, scaling, ceil) +46/-7
Bug fix	dictation.rs Apply dynamic timeout to Whisper/OpenRouter session            openless-all/app/src-tauri/src/coordinator/dictation.rs Use buffer_duration_ms() and whisper_transcribe_timeout() for dynamic timeout Log audio duration and computed timeout duration Update error log message to include actual timeout seconds +12/-4

---

[github-actions]

PR Reviewer Guide 🔍

(Review updated until commit e1f24b6)

Here are some key observations to aid the review process:

🎫 Ticket compliance analysis 🔶 721 - Partially compliant Compliant requirements: Whisper-compatible ASR, especially OpenRouter Whisper path, no longer use fixed 15s global timeout. (Dynamic timeout introduced, global fallback increased to 30s.) OpenRouter Whisper uses dynamic timeout based on audio length. Do not revert feat(asr): 支持 OpenRouter 作为 ASR 提供商 (#582) #591's OpenRouter JSON+base64 request format. Non-compliant requirements: Log audio duration, chunk count, provider, timeout seconds. (Only audio duration and timeout are logged; chunk count and provider are missing.) OpenRouter chunk requests can tolerate brief network jitter; errors distinguish provider HTTP error, network error, client timeout. (No error type differentiation added.) Provide retry path on transcribe failure: preserve audio/PCM for retry or at least allow re-transcribe in debug/settings. (No retry path implemented; existing PCM retention is unchanged but no retry UI/logic.) Requires further human verification: (empty)

⏱️ Estimated effort to review: 1 🔵⚪⚪⚪⚪

🧪 PR contains tests

🔒 No security concerns identified

⚡ No major issues detected

[github-actions]

Persistent review updated to latest commit e1f24b6