feat(transcribe,telegram,config): voice notes, round videos and audio get a transcript from pluggable providers, an openai-compatible one built in

tests/test_transcribe.py

@@ -0,0 +1,101 @@
+import tempfile
+from pathlib import Path
+from typing import Any
+
+import pytest
+from aioresponses import CallbackResult, aioresponses
+
+from beaver_gateway.transcribe import Audio, OpenAITranscriber, transcribe
+
+URL = "http://whisper.local/v1/audio/transcriptions"
+
+
+@pytest.fixture
+def audio() -> Audio:
+    path = Path(tempfile.mkdtemp(prefix="beaver-stt-")) / "u1.ogg"
+    path.write_bytes(b"OggS")
+    return Audio(path, "audio/ogg", 3)
+
+
+def saying(text: str | None, calls: list[str] | None = None):
+    async def transcriber(audio: Audio) -> str | None:
+        if calls is not None:
+            calls.append(text or "")
+        return text
+
+    return transcriber
+
+
+async def broken(audio: Audio) -> str | None:
+    msg = "provider down"
+    raise ConnectionError(msg)
+
+
+async def test_the_first_non_empty_text_wins(audio: Audio) -> None:
+    calls: list[str] = []
+    chain = [saying(None, calls), saying("  \n", calls), saying(" привет ", calls)]
+    chain.append(saying("never", calls))
+    assert await transcribe(chain, audio) == "привет"
+    assert calls == ["", "  \n", " привет "]
+
+
+async def test_a_raising_provider_is_skipped(audio: Audio, caplog) -> None:
+    assert await transcribe([broken, saying("ok")], audio) == "ok"
+    (record,) = [r for r in caplog.records if "provider down" in r.getMessage()]
+    assert record.levelname == "WARNING" and record.exc_info is None
+
+
+async def test_nothing_comes_back_as_none(audio: Audio) -> None:
+    assert await transcribe([], audio) is None
+    assert await transcribe([broken, saying("")], audio) is None
+
+
+async def test_openai_transcriber_posts_the_file_and_reads_the_text(
+    audio: Audio,
+) -> None:
+    seen: dict[str, Any] = {}
+
+    def answer(url, **kwargs: Any) -> CallbackResult:
+        fields = kwargs["data"]._fields
+        seen["headers"] = kwargs["headers"]
+        seen["fields"] = {f[0]["name"]: f[2] for f in fields}
+        seen["file"] = next(f for f in fields if f[0]["name"] == "file")
+        return CallbackResult(payload={"text": " привет "})
+
+    stt = OpenAITranscriber(
+        base_url="http://whisper.local/v1/", api_key="sk", model="m", language="ru"
+    )
+    with aioresponses() as http:
+        http.post(URL, callback=answer)
+        assert await stt(audio) == " привет "
+    assert seen["headers"] == {"Authorization": "Bearer sk"}
+    assert seen["fields"]["model"] == "m"
+    assert seen["fields"]["language"] == "ru"
+    assert seen["fields"]["response_format"] == "json"
+    options, headers, _ = seen["file"]
+    assert options["filename"] == "u1.ogg"
+    assert headers["Content-Type"] == "audio/ogg"
+
+
+async def test_openai_transcriber_without_a_key_sends_no_auth(audio: Audio) -> None:
+    seen: dict[str, Any] = {}
+
+    def answer(url, **kwargs: Any) -> CallbackResult:
+        seen["headers"] = kwargs["headers"]
+        seen["fields"] = [f[0]["name"] for f in kwargs["data"]._fields]
+        return CallbackResult(payload={"text": ""})
+
+    with aioresponses() as http:
+        http.post(URL, callback=answer)
+        assert (
+            await OpenAITranscriber(base_url="http://whisper.local/v1")(audio) is None
+        )
+    assert seen["headers"] == {}
+    assert "language" not in seen["fields"]
+
+
+async def test_openai_transcriber_error_reaches_the_chain(audio: Audio) -> None:
+    stt = OpenAITranscriber(base_url="http://whisper.local/v1")
+    with aioresponses() as http:
+        http.post(URL, status=401, payload={"error": "bad key"})
+        assert await transcribe([stt, saying("fallback")], audio) == "fallback"