fix: recover empty short Whisper transcriptions

[Mr-Sunglasses]

Problem

WhisperKit defaults to excluding one second at the end of a decoding window. Its decode loop runs only while seek < end - windowPadding, so recordings of one second or less never enter the loop and return empty text. A current-source Tiny reproduction returned empty for a 0.498-second synthetic “Hello” recording.

Changes

Set windowClipTime to zero for clips of at most 16,000 samples (one second at 16 kHz). Longer clips retain WhisperKit's default trailing-window protection. Short speech is decoded directly using the original samples and duration, with no additional padding or retry. Existing vocabulary recovery and no-speech filtering remain in place.

Validation

• 14 focused Whisper service tests passed, including decode-loop eligibility at one sample, sub-second lengths, exactly one second, and preservation of default options above the boundary.
• Live Tiny CLI: the original 0.498-second fixture and exactly one-second fixture both returned Hello. with their original durations.
• Digital silence was rejected by the existing CLI audio validation.
• App compiled as part of swift test; git diff --check passed.
• Full suite before self-review: 444 tests, one environment-dependent skip, no failures. The revised code received the focused checks above; CI reruns on the new commit.

[netlify]

✅ Deploy Preview for voca-mac canceled.

Name	Link
🔨 Latest commit	32294c5
🔍 Latest deploy log	https://app.netlify.com/projects/voca-mac/deploys/6a9b60f65be7de00089707d9

[greptile-apps]

Greptile Summary

This PR changes short-clip Whisper transcription to enter the decode loop by removing the trailing exclusion window for audio at or below one second.

• Selects a zero-second clip window for up to 16,000 samples while preserving WhisperKit’s default for longer recordings.
• Adds boundary-focused tests for short and longer sample counts.

Confidence Score: 5/5

The PR appears safe to merge because no blocking failure remains within the eligible follow-up-review scope.

No blocking failure remains.

Important Files Changed

Filename	Overview
Sources/VocaMac/Services/WhisperService.swift	Configures the WhisperKit trailing exclusion window according to the input sample count.
Tests/VocaMacTests/WhisperServiceTests.swift	Adds boundary tests covering decode-loop eligibility and preservation of the longer-clip default.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Whisper audio samples] --> B{Sample count ≤ 16,000?}
    B -->|Yes| C[Set windowClipTime to 0]
    B -->|No| D[Keep one-second window]
    C --> E[Run WhisperKit transcription]
    D --> E

Loading

_{Reviews (2): Last reviewed commit: "fix: decode short Whisper clips without ..." | Re-trigger Greptile}

[jatinkrmalik]

@Mr-Sunglasses thanks, this looks good. Merging.