diegosouzapw/OmniRoute v2.9.2

v2.9.2 — Transcription Content-Type fix, Deepgram language detection, TTS error display

on GitHub

🐛 Bug Fixes

🎤 Audio Transcription

• Content-Type mapping fix — New resolveAudioContentType() helper maps video/mp4 → audio/mp4 so Deepgram correctly processes video files. Previously caused "No speech detected" for all MP4 uploads.
• Auto-language detection — Added detect_language=true to Deepgram requests. Deepgram now auto-detects the audio language (Portuguese, Spanish, French, etc.) instead of defaulting to English.
• Punctuation — Added punctuate=true for higher-quality output from Deepgram Nova 3.
• HuggingFace — Updated to also use resolveAudioContentType() for consistent behavior.

🔊 Text-to-Speech

• [object Object] error fix — upstreamErrorResponse() in audioSpeech.ts and audioTranscription.ts now correctly extracts the string message from nested error objects (e.g., ElevenLabs returns { error: { message: "...", status_code: 401 } }). Users now see readable errors like "Invalid API key" instead of "[object Object]".

📚 Documentation

• README.md: Transcription row updated with provider list (7 providers) and MP4/language-detection capabilities
• README.md: TTS row updated with full provider list (10 providers)
• 29/30 i18n language README files synced
• CHANGELOG.md updated

🔀 Issues Triaged

• #508 → needs-info (tool call regression needs proxy logs)
• #510 → needs-info (Windows CLI path issue)
• #485 → Closed (Kiro external issue)
• #442 → Closed (Baseten /models workaround documented)
• #464 → Roadmap item acknowledged

✅ Tests

• 821/821 tests pass, 0 failures

What's Changed

• chore(release): v2.9.2 — Transcription Content-Type fix, Deepgram language detection, TTS error display by @diegosouzapw in #514

Full Changelog: v2.9.1...v2.9.2