Initial commit: sermon-clean v0.1.0

Find, cut, paste replacements in sermon audio.
- engine.py: SermonClean core (BadSegment, trim, concat, splice verify)
- cli.py: find/cut/paste/pipe subcommands
- elevenlabs.py: optional ElevenLabs rendering hook
- 25 tests passing (timestamp parsing, BadSegment validation, end-to-end trim on real audio)
- pyproject.toml: pip-installable; dep only on ffmpeg/ffprobe system-wide
- README + LICENSE (MIT) + examples/segs.json

Validated against the krystie audio cache (OGG/Opus 48kHz mono).

tests/test_engine.py

@@ -0,0 +1,99 @@
+"""Tests for sermon-clean.engine."""
+
+from pathlib import Path
+
+import pytest
+
+from sermon_clean.engine import (
+    SermonClean,
+    parse_timestamp,
+    format_timestamp,
+)
+
+
+class TestTimestampParsing:
+    @pytest.mark.parametrize("s,expected", [
+        ("30", 30.0),
+        ("90", 90.0),
+        ("1:30", 90.0),
+        ("01:30", 90.0),
+        ("1:30.5", 90.5),
+        ("1:00:30", 3630.0),
+        ("02:01:30.5", 7290.5),
+    ])
+    def test_valid(self, s, expected):
+        assert parse_timestamp(s) == expected
+
+    @pytest.mark.parametrize("s", ["", "abc", "1:60", "60:60", "1.2.3"])
+    def test_invalid(self, s):
+        with pytest.raises(ValueError):
+            parse_timestamp(s)
+
+
+class TestFormatTimestamp:
+    @pytest.mark.parametrize("seconds,expected", [
+        (0.0, "00:00.000"),
+        (59.999, "00:59.999"),
+        (60.0, "01:00.000"),
+        (90.5, "01:30.500"),
+        (3600.0, "01:00:00.000"),
+        (3661.5, "01:01:01.500"),
+        (7290.5, "02:01:30.500"),
+    ])
+    def test_round_trip(self, seconds, expected):
+        assert format_timestamp(seconds) == expected
+
+    def test_roundtrip_parse(self):
+        for s in [0.0, 5.0, 59.999, 60.0, 3630.0, 7290.5]:
+            assert parse_timestamp(format_timestamp(s)) == pytest.approx(s)
+
+
+class TestBadSegment:
+    def test_valid(self):
+        from sermon_clean.engine import BadSegment
+        s = BadSegment(10.0, 12.0, reason="misspoke")
+        assert s.duration == 2.0
+        assert s.start == 10.0
+        assert s.end == 12.0
+
+    def test_zero_length_rejected(self):
+        from sermon_clean.engine import BadSegment
+        with pytest.raises(ValueError):
+            BadSegment(10.0, 10.0)
+
+    def test_negative_rejected(self):
+        from sermon_clean.engine import BadSegment
+        with pytest.raises(ValueError):
+            BadSegment(15.0, 10.0)
+
+    def test_long_window_rejected(self):
+        from sermon_clean.engine import BadSegment
+        # 90s window by default > 60s cap
+        with pytest.raises(ValueError, match="suspiciously long"):
+            BadSegment(0.0, 90.0)
+
+
+def test_sermon_clean_basic(tmp_path):
+    """Smoke test: load a real audio file from the krystie profile cache."""
+    # Find any audio file in the krystie cache
+    cache = Path("/root/.hermes/profiles/krystie/cache/audio")
+    if not cache.exists():
+        pytest.skip("krystie audio cache not present")
+    audio_files = sorted(cache.glob("*.ogg")) + sorted(cache.glob("*.mp3"))
+    if not audio_files:
+        pytest.skip("no audio files in krystie cache")
+    audio = audio_files[0]
+    sc = SermonClean(audio, workdir=tmp_path)
+    assert sc.duration > 0
+    assert sc.codec["codec_name"]
+    assert sc.codec["sample_rate"]
+    seg = sc.add_segment(0.5, 1.0, reason="test")
+    assert seg.start == 0.5
+    assert len(sc.segments) == 1
+    segs = sc.trim_segments()
+    assert len(segs) == 2  # before + after the 0.5s window
+    assert all(p.exists() for p in segs)
+    # Verify the trimmed segments
+    from sermon_clean.engine import probe_duration
+    total = sum(probe_duration(p) for p in segs)
+    assert total == pytest.approx(sc.duration - 0.5, abs=0.1)