Add speaker-attributed meeting transcription

tests/test_recording.py

@@ -0,0 +1,94 @@
+import asyncio
+import json
+import os
+import sys
+import wave
+from pathlib import Path
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(__file__)))
+
+import recording
+
+
+class FakeUser:
+    def __init__(self, user_id, display_name=None, name=None):
+        self.id = user_id
+        if display_name is not None:
+            self.display_name = display_name
+        if name is not None:
+            self.name = name
+
+
+class FakeData:
+    def __init__(self, pcm):
+        self.pcm = pcm
+
+
+def _read_wav_frames(path):
+    with wave.open(str(path), "rb") as w:
+        return w.readframes(w.getnframes())
+
+
+SPEECH = b"\x10\x20" * 100  # non-silent PCM
+SILENCE = b"\x00\x00" * 100
+
+
+def test_meeting_sink_writes_mixed_and_per_speaker_tracks(tmp_path):
+    mixed = tmp_path / "meeting.wav"
+    tracks_dir = tmp_path / "tracks"
+    sink = recording.MeetingSink(str(mixed), str(tracks_dir))
+
+    alice = FakeUser(111, display_name="Alice")
+    bob = FakeUser(222, display_name="Bob")
+
+    sink.write(alice, FakeData(SPEECH))
+    sink.write(bob, FakeData(SPEECH))
+    sink.write(alice, FakeData(SPEECH))
+    sink.cleanup()
+
+    # Mixed archive is a valid, finalized WAV containing all frames.
+    assert mixed.exists()
+    mixed_frames = _read_wav_frames(mixed)
+    assert mixed_frames == SPEECH * 3
+
+    # Each speaker has their own track.
+    alice_track = tracks_dir / "track-111.wav"
+    bob_track = tracks_dir / "track-222.wav"
+    assert _read_wav_frames(alice_track) == SPEECH * 2
+    assert _read_wav_frames(bob_track) == SPEECH
+
+    # Manifest maps tracks back to display names.
+    manifest = json.loads((tracks_dir / "tracks.json").read_text())
+    names = {t["display_name"]: t["user_id"] for t in manifest["tracks"]}
+    assert names == {"Alice": 111, "Bob": 222}
+
+
+def test_meeting_sink_excludes_silence_from_speaker_tracks(tmp_path):
+    mixed = tmp_path / "meeting.wav"
+    tracks_dir = tmp_path / "tracks"
+    sink = recording.MeetingSink(str(mixed), str(tracks_dir))
+
+    alice = FakeUser(111, display_name="Alice")
+    sink.write(alice, FakeData(SPEECH))
+    sink.write(alice, FakeData(SILENCE))  # synthetic silence frame
+    sink.cleanup()
+
+    # Mixed gets both frames (faithful timeline), speaker track only the speech.
+    assert _read_wav_frames(mixed) == SPEECH + SILENCE
+    assert _read_wav_frames(tracks_dir / "track-111.wav") == SPEECH
+
+
+def test_meeting_sink_ignores_silence_only_and_unknown_users(tmp_path):
+    mixed = tmp_path / "meeting.wav"
+    tracks_dir = tmp_path / "tracks"
+    sink = recording.MeetingSink(str(mixed), str(tracks_dir))
+
+    # Silence with no user, then silence with a user: neither creates a track.
+    sink.write(None, FakeData(SILENCE))
+    sink.write(FakeUser(999, display_name="Quiet"), FakeData(SILENCE))
+    sink.cleanup()
+
+    manifest = json.loads((tracks_dir / "tracks.json").read_text())
+    assert manifest["tracks"] == []
+    # Mixed archive still finalized and valid.
+    assert _read_wav_frames(mixed) == SILENCE * 2