smshine/mediaplayer
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add debug info

Browse Source
This commit is contained in:
Shuming Liu
2026-10-25 11:31:32 +08:00
parent ae84347912
commit 5428c42ffd
5 changed files with 212 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

107
sentence_api/processing.py
View File

@@ -200,15 +200,50 @@ class VideoProcessor:
gaps = find_sentence_gaps(document, min_gap_ms)
inserted: List[SentenceBoundary] = []
transcribed_gaps = 0
logger.info(
"Gap repair started: video=%s min_gap_ms=%d language=%s gaps=%d "
"sentences=%d duration_ms=%d transcriber=%s",
video_hash,
min_gap_ms,
video.get("language") or "auto",
len(gaps),
len(document.sentences),
document.duration_ms,
type(self.transcriber).__name__,
)
if gaps:
language = video.get("language")
with tempfile.TemporaryDirectory(prefix="gap-repair-") as temp_dir:
for gap_index, (raw_start, raw_end) in enumerate(gaps):
logger.info(
"Gap %d/%d: raw=%d-%dms duration=%dms",
gap_index + 1,
len(gaps),
raw_start,
raw_end,
raw_end - raw_start,
)
clipped = _inset_gap(raw_start, raw_end, document.duration_ms)
if clipped is None:
logger.warning(
"Gap %d/%d skipped after inset: raw=%d-%dms",
gap_index + 1,
len(gaps),
raw_start,
raw_end,
)
continue
start_ms, end_ms = clipped
chunk_path = Path(temp_dir) / f"gap-{gap_index}.wav"
logger.info(
"Gap %d/%d extracted range=%d-%dms duration=%dms file=%s",
gap_index + 1,
len(gaps),
start_ms,
end_ms,
end_ms - start_ms,
chunk_path,
)
try:
extract_audio_segment(
media_path,
@@ -219,14 +254,33 @@ class VideoProcessor:
samples, sample_rate = decode_audio_mono(chunk_path)
metrics = analyze_samples(samples, sample_rate)
except (AudioAnalysisError, RuntimeError) as exc:
logger.info(
logger.warning(
"Skipping sentence gap %s-%sms: %s",
start_ms,
end_ms,
exc,
exc_info=True,
)
continue
logger.info(
"Gap %d/%d audio metrics: duration_ms=%d speech_duration_ms=%d "
"internal_silence_ms=%d pause_ratio=%.3f threshold_ms=%d",
gap_index + 1,
len(gaps),
metrics.recording_duration_ms,
metrics.speech_duration_ms,
metrics.internal_silence_ms,
metrics.internal_pause_ratio,
GAP_REPAIR_MIN_SPEECH_MS,
)
if metrics.speech_duration_ms < GAP_REPAIR_MIN_SPEECH_MS:
logger.warning(
"Gap %d/%d skipped as silence: speech_duration_ms=%d < %d",
gap_index + 1,
len(gaps),
metrics.speech_duration_ms,
GAP_REPAIR_MIN_SPEECH_MS,
)
continue
try:
transcript = self.transcriber.transcribe(chunk_path, language)
@@ -237,8 +291,16 @@ class VideoProcessor:
start_ms,
end_ms,
exc,
exc_info=True,
)
continue
logger.info(
"Gap %d/%d transcript returned: segments=%d text=%r",
gap_index + 1,
len(gaps),
len(transcript.segments),
transcript.text,
)
gap_document = document_from_transcript(
video_hash=video_hash,
duration_ms=end_ms - start_ms,
@@ -252,12 +314,45 @@ class VideoProcessor:
offset_ms=start_ms,
limit_end_ms=end_ms,
)
if not shifted:
logger.warning(
"Gap %d/%d produced no insertable sentences: "
"transcript_segments=%d boundary_sentences=%d range=%d-%dms",
gap_index + 1,
len(gaps),
len(transcript.segments),
len(gap_document.sentences),
start_ms,
end_ms,
)
for boundary in gap_document.sentences:
logger.warning(
"Candidate boundary before shift: start=%d end=%d text=%r",
boundary.start_ms,
boundary.end_ms,
boundary.text,
)
if not shifted:
continue
transcribed_gaps += 1
inserted.extend(shifted)
logger.info(
"Gap %d/%d produced %d insertable sentences: range=%d-%dms",
gap_index + 1,
len(gaps),
len(shifted),
start_ms,
end_ms,
)
if not inserted:
logger.warning(
"Gap repair finished without inserts: video=%s gaps=%d "
"transcribed_gaps=%d inserted=0",
video_hash,
len(gaps),
transcribed_gaps,
)
return {
"min_gap_ms": min_gap_ms,
"gap_count": len(gaps),
@@ -269,6 +364,16 @@ class VideoProcessor:
merged = _merge_sentence_lists(document.sentences, inserted)
updated = document.model_copy(update={"sentences": merged})
saved = self.repository.replace_sentences(updated)
logger.info(
"Gap repair finished: video=%s gaps=%d transcribed_gaps=%d inserted=%d "
"sentences=%d->%d",
video_hash,
len(gaps),
transcribed_gaps,
len(saved.sentences) - len(document.sentences),
len(document.sentences),
len(saved.sentences),
)
return {
"min_gap_ms": min_gap_ms,
"gap_count": len(gaps),