Merge pull request chigwell#229 from omid-io/fix/groq-oga-audio-transcription

fix(transcription): normalize oga audio extension to ogg for Groq uploads

Author: chigwell

telegram_mcp/transcription.py

@@ -362,7 +362,11 @@ async def _transcribe_via_groq(cl, msg) -> dict:
         return _too_large_for_groq(size, limit)
 
     file_attr = getattr(msg, "file", None)
-    ext = (getattr(file_attr, "ext", None) or ".oga").lstrip(".")
+    ext = (getattr(file_attr, "ext", None) or "ogg").lstrip(".")
+    # Telegram voice notes carry 'oga' (Opus-in-Ogg), which Groq rejects with 400
+    # unsupported_audio_format. Groq requires .ogg, so normalize oga -> ogg.
+    if ext.lower() == "oga":
+        ext = "ogg"
     mime = getattr(file_attr, "mime_type", None) or "audio/ogg"
 
     try:

tests/test_transcription.py

@@ -446,10 +446,53 @@ async def _download_media(m, file=None):
 
     assert result == {"status": "ok", "text": "hello world", "lang": "ru"}
     assert calls[0]["headers"]["Authorization"] == "Bearer test-key"
-    assert calls[0]["files"]["file"][0] == "voice.oga"
+    assert calls[0]["files"]["file"][0] == "voice.ogg"
     assert calls[0]["data"]["model"] == transcription.GROQ_MODEL
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    "ext_in,expected_filename",
+    [
+        (".oga", "voice.ogg"),
+        ("oga", "voice.ogg"),
+        (".ogg", "voice.ogg"),
+        (".mp3", "voice.mp3"),
+        (".wav", "voice.wav"),
+        (None, "voice.ogg"),
+    ],
+)
+async def test_transcribe_via_groq_normalizes_audio_file_extension(
+    monkeypatch, ext_in, expected_filename
+):
+    """Groq API audio transcription only accepts [flac mp3 mp4 mpeg mpga m4a ogg opus wav webm].
+    Telegram voice notes arrive as .oga, which Groq rejects with 400 unsupported_audio_format.
+    Naming the upload as .ogg satisfies Groq while preserving the identical audio content (issue #219).
+    """
+    monkeypatch.setenv("GROQ_API_KEY", "test-key")
+    calls = []
+    response = _FakeGroqResponse({"text": "ok", "language": "en"})
+    monkeypatch.setattr(
+        transcription.httpx, "AsyncClient", lambda **kw: _FakeHttpxClient(response, calls)
+    )
+
+    file_obj = (
+        SimpleNamespace(duration=10, ext=ext_in, mime_type="audio/ogg")
+        if ext_in is not None
+        else None
+    )
+    msg = _voice_msg(file=file_obj)
+
+    async def _download_media(m, file=None):
+        return b"audio-bytes"
+
+    client = SimpleNamespace(download_media=_download_media)
+    result = await transcription._transcribe_via_groq(client, msg)
+
+    assert result["status"] == "ok"
+    assert calls[0]["files"]["file"][0] == expected_filename
+
+
 @pytest.mark.asyncio
 async def test_transcribe_via_groq_requires_api_key(monkeypatch):
     monkeypatch.delenv("GROQ_API_KEY", raising=False)