fix(stt): accept .oga and .opus voice notes for transcription

[eliemada]

What does this PR do?

Telegram delivers voice notes as .oga (OGG/Opus container). SUPPORTED_FORMATS in tools/transcription_tools.py listed .ogg but not .oga, so transcribe_audio() rejected every inbound Telegram voice note with Unsupported format: .oga before it reached any STT backend. The gateway then replied "I couldn't transcribe your voice message."

The check sits upstream of provider dispatch, so this broke voice notes for all backends (local faster-whisper, Groq, OpenAI, Mistral, xAI) and any command-provider. The fix adds .oga and .opus to the allowlist.

Related Issue

No existing issue. Bug and reproduction below.

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• tools/transcription_tools.py: add .oga and .opus to SUPPORTED_FORMATS.
• tests/tools/test_transcription_tools.py: add test_telegram_oga_and_opus_accepted regression test.

How to Test

Reproduce (before the fix): send a voice note to a Telegram gateway, or call transcribe_audio("voice.oga") — it returns {"success": false, "error": "Unsupported format: .oga ..."}.

After the fix:

pytest tests/tools/test_transcription_tools.py -q
# 108 passed

_validate_audio_file() now returns None (accepted) for .oga and .opus. Verified end to end on Ubuntu 24.04 with a real Telegram voice note routed to a command-provider STT backend.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix
• I've run pytest tests/tools/test_transcription_tools.py -q and all 108 tests pass
• I've added a regression test for this change
• I've tested on my platform: Ubuntu 24.04

Documentation & Housekeeping

• No docs change needed — N/A
• No config keys changed — N/A
• No architecture/workflow change — N/A
• Cross-platform: extension allowlist, no platform-specific code — N/A

[teknium1]

Thanks for the focused regression test. Current main still rejects .opus in tools/transcription_tools.py:103, and Feishu can retain .opus when caching inbound audio (plugins/platforms/feishu/adapter.py:172,3925-3929).

Problems

• The reported Telegram path is no longer reproducible on current main: inbound msg.voice is explicitly cached as .ogg at plugins/platforms/telegram/adapter.py:7849-7853.
• Please do not make .oga a bare shared allowlist entry without reconciling the current filename-normalization contract. WhatsApp maps Ogg/Opus MIME values to .ogg because .oga is documented as downstream-STT-incompatible (gateway/platforms/whatsapp_cloud.py:129-164), while several cloud handlers upload the original basename (tools/transcription_tools.py:1406-1409, 1479-1488, 1566-1571).

Suggested changes

• Re-scope around the verified .opus ingress or canonicalize Ogg/Opus variants before remote dispatch.
• Add a regression through the relevant cache-to-STT path, not only _validate_audio_file().

Automated hermes-sweeper review.

[teknium1]

Current Telegram inbound voice handling already forces .ogg (plugins/platforms/telegram/adapter.py:7849-7853), while WhatsApp deliberately maps .oga to .ogg because downstream STT expects that filename (gateway/platforms/whatsapp_cloud.py:129-164). Please establish provider-safe canonicalization before adding .oga to the shared allowlist; this change otherwise forwards the .oga basename unchanged to cloud providers.

[teknium1]

Merged into main via consolidated salvage PR #73507 (merge 156edc5ded). Your .oga/.opus additions to the STT SUPPORTED_FORMATS (Telegram voice notes are .oga) were cherry-picked as 2447d2e.

Your contribution is credited to you in git history. Thank you! Closing this PR as merged-via-salvage.