Fix: live recording + whisper.cpp + Translate task → "File Not Found (/audio/translations)"

[NumerousDragonfly]

Fix: live recording + whisper.cpp + Translate task → "File Not Found (/audio/translations)"

Bug

Using live recording with the whisper.cpp backend, source language zh,
task Translate, fails immediately with:
Transcription (task=Transcribe) works fine on the same setup; only
Translate is broken.

Root cause

start_local_whisper_server() boots the bundled whisper-server with a
single hardcoded route:

"--inference-path", "/audio/transcriptions",

Meanwhile, _transcribe_via_api() picks between the OpenAI SDK's
.audio.transcriptions.create() and .audio.translations.create()
based on task. When task is Translate, the SDK POSTs to
/audio/translations — a route the local server never registers — so
the request 404s.

This affects every user running live recording + whisper.cpp + Translate,
not just a specific machine/config. File-based transcription (whisper_cpp.py,
which invokes whisper-cli directly with a --translate flag) is unaffected.

Fix

whisper.cpp's whisper-server supports its own --translate flag, which
makes its single endpoint return translated text instead of a same-language
transcript. Two small changes:

1. start_local_whisper_server() passes --translate to the server
   subprocess when the selected task is Translate.
2. _transcribe_via_api() always calls .audio.transcriptions.create()
   when the backend is the local whisper.cpp server, since that's the
   only route it exposes — regardless of task.

The OpenAI Whisper API backend (ModelType.OPEN_AI_WHISPER_API), which
genuinely exposes both /audio/transcriptions and /audio/translations
on the real OpenAI API, is unaffected — the transcribe/translate branch
there is preserved exactly as before.

Testing

• Live recording, whisper.cpp backend, task=Transcribe, zh → unchanged, returns Chinese text
• Live recording, whisper.cpp backend, task=Translate, zh → now returns English text, no 404
• Live recording/API mode, OpenAI Whisper API backend, task=Transcribe → unchanged
• Live recording/API mode, OpenAI Whisper API backend, task=Translate → unchanged
• File-based transcription (whisper.cpp, Translate) → unaffected, still uses whisper-cli directly

Notes for reviewers

No new endpoints, no changes to file-based transcription, no changes to
faster-whisper/Hugging Face code paths. Scope is limited to the
live-recording + local-whisper.cpp-server code path.

[raivisdejus]

@NumerousDragonfly Thanks for your contribution! Nice fix

[codecov]

Codecov Report

❌ Patch coverage is 75.00000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 82.95%. Comparing base (191795b) to head (1fccefe).

Files with missing lines	Patch %	Lines
buzz/transcriber/recording_transcriber.py	75.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1555      +/-   ##
==========================================
- Coverage   82.98%   82.95%   -0.03%     
==========================================
  Files         108      108              
  Lines       11693    11697       +4     
==========================================
  Hits         9703     9703              
- Misses       1990     1994       +4     

Flag	Coverage Δ
Linux	81.71% <75.00%> (-0.02%)	⬇️
macOS	82.46% <75.00%> (-0.03%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.