Fix voice transcription handling

backend/modal_api.py

@@ -19,15 +19,15 @@ REQUEST_TIMEOUT_SECONDS = 120
 
 
 def transcribe_audio(
-    audio_path: str | Path | None,
-    fallback: Callable[[str | Path | None], str],
+    audio_path: Any,
+    fallback: Callable[[Any], str],
 ) -> str:
     """Transcribe audio through Modal, falling back locally if unavailable."""
+    path = _coerce_audio_path(audio_path)
     endpoint_url = os.getenv(MODAL_TRANSCRIBE_URL_ENV)
-    if not endpoint_url or audio_path is None:
+    if not endpoint_url or path is None:
         return fallback(audio_path)
 
-    path = Path(audio_path)
     if not path.exists():
         return fallback(audio_path)
 
@@ -79,3 +79,22 @@ def _auth_headers() -> dict[str, str]:
     if not token:
         return {}
     return {"Authorization": f"Bearer {token}"}
+
+
+def _coerce_audio_path(audio_value: Any) -> Path | None:
+    """Extract a local audio filepath from Gradio audio values."""
+    if audio_value is None:
+        return None
+    if isinstance(audio_value, (str, Path)):
+        return Path(audio_value)
+    if isinstance(audio_value, dict):
+        for key in ("path", "name", "file", "filepath"):
+            value = audio_value.get(key)
+            if value:
+                return Path(value)
+    if isinstance(audio_value, (list, tuple)):
+        for value in audio_value:
+            path = _coerce_audio_path(value)
+            if path is not None:
+                return path
+    return None

tests/test_modal_api.py

@@ -82,6 +82,21 @@ def test_transcribe_audio_uses_modal_response(tmp_path: Path, monkeypatch) -> No
     assert transcript == "Sold 12 mangoes"
 
 
+def test_transcribe_audio_accepts_gradio_dict_payload(tmp_path: Path, monkeypatch) -> None:
+    audio_path = tmp_path / "audio.wav"
+    audio_path.write_bytes(b"audio")
+    monkeypatch.setenv(modal_api.MODAL_TRANSCRIBE_URL_ENV, "https://modal.example/transcribe")
+
+    def fake_post(*args, **kwargs) -> FakeResponse:
+        return FakeResponse({"transcript": "Paid 500 for supplies"})
+
+    monkeypatch.setattr(modal_api.requests, "post", fake_post)
+
+    transcript = modal_api.transcribe_audio({"path": str(audio_path)}, fallback=lambda _: "fallback")
+
+    assert transcript == "Paid 500 for supplies"
+
+
 def test_transcribe_audio_falls_back_when_url_missing(tmp_path: Path, monkeypatch) -> None:
     audio_path = tmp_path / "audio.wav"
     audio_path.write_bytes(b"audio")

tests/test_speech.py

@@ -15,3 +15,10 @@ def test_transcribe_audio_requires_existing_file(tmp_path: Path) -> None:
 
     with pytest.raises(TranscriptionError, match="does not exist"):
         transcribe_audio(missing_file)
+
+
+def test_transcribe_audio_accepts_gradio_dict_for_missing_file(tmp_path: Path) -> None:
+    missing_file = tmp_path / "missing.wav"
+
+    with pytest.raises(TranscriptionError, match="does not exist"):
+        transcribe_audio({"path": str(missing_file)})

tests/test_ui_audio_flow.py

@@ -6,8 +6,8 @@ from voiceledger.ui import gradio_app
 
 
 def test_transcribe_and_parse_audio_handles_transcription_error(monkeypatch) -> None:
-    def fail_transcription(_: str | None) -> str:
-        raise gradio_app.TranscriptionError("test failure")
+    def fail_transcription(_: object) -> str:
+        raise RuntimeError("test failure")
 
     monkeypatch.setattr(gradio_app.modal_api, "transcribe_audio", lambda audio, fallback: fail_transcription(audio))
 

voiceledger/speech/transcribe.py

@@ -5,7 +5,7 @@ from __future__ import annotations
 from functools import lru_cache
 from pathlib import Path
 from collections.abc import Iterable
-from typing import Protocol
+from typing import Any, Protocol
 
 
 MODEL_SIZE = "small"
@@ -28,17 +28,17 @@ class TranscriptionError(RuntimeError):
     """Raised when audio transcription fails."""
 
 
-def transcribe_audio(audio_path: str | Path | None) -> str:
+def transcribe_audio(audio_path: Any) -> str:
     """Transcribe an audio file with the faster-whisper small model.
 
     The model is loaded lazily and cached after the first call. This keeps the
     Gradio app responsive at startup while still using the requested small
     open-source speech model for actual transcription.
     """
-    if audio_path is None:
+    path = _coerce_audio_path(audio_path)
+    if path is None:
         raise TranscriptionError("No audio file was provided.")
 
-    path = Path(audio_path)
     if not path.exists():
         raise TranscriptionError(f"Audio file does not exist: {path}")
 
@@ -62,6 +62,25 @@ def transcribe_audio(audio_path: str | Path | None) -> str:
     return transcript.strip()
 
 
+def _coerce_audio_path(audio_value: Any) -> Path | None:
+    """Extract a local audio filepath from Gradio audio values."""
+    if audio_value is None:
+        return None
+    if isinstance(audio_value, (str, Path)):
+        return Path(audio_value)
+    if isinstance(audio_value, dict):
+        for key in ("path", "name", "file", "filepath"):
+            value = audio_value.get(key)
+            if value:
+                return Path(value)
+    if isinstance(audio_value, (list, tuple)):
+        for value in audio_value:
+            path = _coerce_audio_path(value)
+            if path is not None:
+                return path
+    return None
+
+
 @lru_cache(maxsize=1)
 def _get_model() -> WhisperModelProtocol:
     """Load and cache the faster-whisper small model."""

voiceledger/ui/gradio_app.py

@@ -299,11 +299,11 @@ def _parse_note(note: str) -> tuple[dict[str, Any], dict[str, Any], str]:
     return payload, payload, status
 
 
-def _transcribe_and_parse_audio(audio_path: str | None) -> tuple[str, dict[str, Any], dict[str, Any] | None, str]:
+def _transcribe_and_parse_audio(audio_path: Any) -> tuple[str, dict[str, Any], dict[str, Any] | None, str]:
     """Transcribe recorded audio, parse the transcript, and return UI updates."""
     try:
         transcript = modal_api.transcribe_audio(audio_path, fallback=local_transcribe_audio)
-    except TranscriptionError as exc:
+    except Exception as exc:
         empty_payload = _empty_transaction_payload()
         return "", empty_payload, None, f"Transcription failed: {exc}"