JMR-devandClaude Opus 5
56a54c2b9a
normalize_segments repairs what Whisper actually emits: an end timestamp of -1 when the closing token is never predicted (usually the trailing chunk), out-of-bounds and inverted spans, and blank text. Left alone these produce malformed subtitles. It also warns when chunk starts are non-monotonic, which is the cheap signal that start_ts is window-relative rather than absolute -- that would misplace every cue past 0:30, and it should surface as a log line rather than a user report. Timestamps convert to integer milliseconds before splitting into fields; formatting the seconds field directly renders 3599.9996 as "00:59:60.000". Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
143 lines
4.7 KiB
Python
143 lines
4.7 KiB
Python
from __future__ import annotations
|
|
|
|
import json
|
|
|
|
import pytest
|
|
|
|
from ccn_transcribe import formats
|
|
from ccn_transcribe.transcript import Segment, TranscriptResult
|
|
|
|
SEGMENTS = (Segment(0.0, 1.5, "Hello there."), Segment(1.5, 3.25, "General Kenobi."))
|
|
|
|
RESULT = TranscriptResult(
|
|
segments=SEGMENTS,
|
|
language="en",
|
|
backend="openvino",
|
|
device="GPU",
|
|
model_id="OpenVINO/whisper-large-v3-turbo-int8-ov",
|
|
model_revision="b568445d",
|
|
audio_s=3.25,
|
|
wall_s=1.0,
|
|
)
|
|
|
|
|
|
class TestTimestamps:
|
|
@pytest.mark.parametrize(
|
|
("seconds", "expected"),
|
|
[
|
|
(0.0, "00:00:00,000"),
|
|
(1.5, "00:00:01,500"),
|
|
(61.25, "00:01:01,250"),
|
|
(3600.0, "01:00:00,000"),
|
|
(7322.5, "02:02:02,500"),
|
|
],
|
|
)
|
|
def test_srt_timestamp(self, seconds: float, expected: str) -> None:
|
|
assert formats.srt_timestamp(seconds) == expected
|
|
|
|
def test_vtt_uses_a_dot_separator(self) -> None:
|
|
assert formats.vtt_timestamp(61.25) == "00:01:01.250"
|
|
|
|
def test_rounding_never_produces_sixty_seconds(self) -> None:
|
|
# Naive f"{s:06.3f}" renders this as 00:59:60.000, which is invalid and
|
|
# which some players silently mis-seek on.
|
|
assert formats.srt_timestamp(3599.9996) == "01:00:00,000"
|
|
assert formats.vtt_timestamp(3599.9996) == "01:00:00.000"
|
|
|
|
def test_negative_times_clamp_to_zero(self) -> None:
|
|
assert formats.srt_timestamp(-5.0) == "00:00:00,000"
|
|
|
|
def test_hours_beyond_two_digits_still_render(self) -> None:
|
|
assert formats.srt_timestamp(360000.0).startswith("100:00:00")
|
|
|
|
|
|
class TestTxt:
|
|
def test_joins_segment_text(self) -> None:
|
|
assert formats.to_txt(RESULT) == "Hello there. General Kenobi."
|
|
|
|
def test_ends_with_a_newline(self) -> None:
|
|
assert formats.to_txt(RESULT).endswith("\n") is False # body only; writer adds newline
|
|
|
|
|
|
class TestSrt:
|
|
def test_structure(self) -> None:
|
|
assert formats.to_srt(SEGMENTS) == (
|
|
"1\n"
|
|
"00:00:00,000 --> 00:00:01,500\n"
|
|
"Hello there.\n"
|
|
"\n"
|
|
"2\n"
|
|
"00:00:01,500 --> 00:00:03,250\n"
|
|
"General Kenobi.\n"
|
|
)
|
|
|
|
def test_empty_segments_give_empty_output(self) -> None:
|
|
assert formats.to_srt(()) == ""
|
|
|
|
|
|
class TestVtt:
|
|
def test_has_the_required_header(self) -> None:
|
|
assert formats.to_vtt(SEGMENTS).startswith("WEBVTT\n\n")
|
|
|
|
def test_uses_dot_separated_timestamps_and_no_cue_numbers(self) -> None:
|
|
body = formats.to_vtt(SEGMENTS)
|
|
assert "00:00:00.000 --> 00:00:01.500" in body
|
|
assert "," not in body.split("Hello", maxsplit=1)[0]
|
|
|
|
def test_empty_segments_still_emit_the_header(self) -> None:
|
|
assert formats.to_vtt(()) == "WEBVTT\n"
|
|
|
|
|
|
class TestJson:
|
|
def test_records_run_metadata(self) -> None:
|
|
doc = json.loads(formats.to_json(RESULT))
|
|
assert doc["backend"] == "openvino"
|
|
assert doc["device"] == "GPU"
|
|
assert doc["model"]["id"] == "OpenVINO/whisper-large-v3-turbo-int8-ov"
|
|
assert doc["model"]["revision"] == "b568445d"
|
|
assert doc["language"] == "en"
|
|
assert doc["audio_s"] == 3.25
|
|
assert doc["wall_s"] == 1.0
|
|
assert doc["rtf"] == pytest.approx(3.25)
|
|
|
|
def test_records_segments_with_times(self) -> None:
|
|
doc = json.loads(formats.to_json(RESULT))
|
|
assert doc["segments"] == [
|
|
{"start": 0.0, "end": 1.5, "text": "Hello there."},
|
|
{"start": 1.5, "end": 3.25, "text": "General Kenobi."},
|
|
]
|
|
|
|
def test_includes_full_text(self) -> None:
|
|
assert json.loads(formats.to_json(RESULT))["text"] == "Hello there. General Kenobi."
|
|
|
|
def test_merges_source_metadata_when_given(self) -> None:
|
|
doc = json.loads(formats.to_json(RESULT, source={"id": "abc", "title": "T"}))
|
|
assert doc["source"] == {"id": "abc", "title": "T"}
|
|
|
|
def test_source_key_absent_when_not_given(self) -> None:
|
|
assert "source" not in json.loads(formats.to_json(RESULT))
|
|
|
|
def test_is_utf8_readable_not_escaped(self) -> None:
|
|
result = TranscriptResult(
|
|
segments=(Segment(0.0, 1.0, "naïve café"),),
|
|
language="fr",
|
|
backend="openvino",
|
|
device="CPU",
|
|
model_id="m",
|
|
model_revision=None,
|
|
audio_s=1.0,
|
|
wall_s=1.0,
|
|
)
|
|
assert "naïve café" in formats.to_json(result)
|
|
|
|
|
|
def test_render_returns_every_requested_format() -> None:
|
|
out = formats.render(RESULT, ("txt", "srt", "vtt", "json"))
|
|
assert set(out) == {"txt", "srt", "vtt", "json"}
|
|
assert out["srt"].startswith("1\n")
|
|
|
|
|
|
def test_render_rejects_an_unknown_format() -> None:
|
|
with pytest.raises(ValueError, match="unknown"):
|
|
formats.render(RESULT, ("txt", "docx"))
|