feat: transcription module — faster-whisper with ROCm auto-detect

tests/test_transcription.py

@@ -0,0 +1,25 @@
+import asyncio
+from unittest.mock import MagicMock
+
+
+def test_transcription_engine_is_singleton():
+    from transcription import engine, TranscriptionEngine
+    assert isinstance(engine, TranscriptionEngine)
+
+
+def test_transcribe_file_calls_whisper(tmp_path):
+    wav = tmp_path / "test.wav"
+    wav.write_bytes(b"\x00" * 100)
+
+    mock_model = MagicMock()
+    mock_segment = MagicMock()
+    mock_segment.text = " Hallo Welt"
+    mock_model.transcribe.return_value = ([mock_segment], MagicMock())
+
+    from transcription import TranscriptionEngine
+    eng = TranscriptionEngine()
+    eng._model = mock_model
+
+    result = asyncio.run(eng.transcribe_file(str(wav), language="de"))
+    assert result == "Hallo Welt"
+    mock_model.transcribe.assert_called_once_with(str(wav), language="de")

transcription.py

@@ -0,0 +1,36 @@
+import asyncio
+
+
+class TranscriptionEngine:
+    _model = None
+
+    def _get_model(self, model_name: str = "large-v3", device: str = "auto"):
+        if self._model is None:
+            from faster_whisper import WhisperModel
+            if device == "auto":
+                try:
+                    self._model = WhisperModel(model_name, device="cuda", compute_type="float16")
+                except Exception:
+                    self._model = WhisperModel(model_name, device="cpu", compute_type="int8")
+            else:
+                compute = "float16" if device in ("cuda", "rocm") else "int8"
+                self._model = WhisperModel(model_name, device=device, compute_type=compute)
+        return self._model
+
+    async def transcribe_file(
+        self,
+        audio_path: str,
+        language: str = "de",
+        model_name: str = "large-v3",
+        device: str = "auto",
+    ) -> str:
+        loop = asyncio.get_event_loop()
+        model = self._get_model(model_name, device)
+        segments, _ = await loop.run_in_executor(
+            None,
+            lambda: model.transcribe(audio_path, language=language),
+        )
+        return "".join(seg.text for seg in segments).strip()
+
+
+engine = TranscriptionEngine()