kura/crosssubtitle-ai
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

避免重复的 segment

Browse Source
This commit is contained in:
kura 2026-04-30 15:47:58 +08:00
parent a7046eba8c
commit 479049501b
1 changed files with 13 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
src-tauri/src/whisper.rs
View File

@ -104,9 +104,9 @@ impl WhisperEngine {
let vad_coverage = speech_coverage_ratio(&normalized_ranges, total_seconds);
let should_retry_full_audio = !audio.is_empty()
&& (segments.is_empty()
|| vad_coverage < 0.72
|| vad_end + 2.5 < total_seconds
|| (total_seconds > 45.0 && vad_text_len < (total_seconds / 2.4) as usize));
|| vad_coverage < 0.60
|| vad_end + 5.0 < total_seconds
|| (total_seconds > 60.0 && vad_text_len < (total_seconds / 3.0) as usize));
if should_retry_full_audio {
on_log(format!(
@ -141,6 +141,12 @@ impl WhisperEngine {
))?;
segments = full_audio_segments;
} else {
on_log(format!(
"whisper: keeping VAD-based transcript (vad_segments={}, full_segments={})",
segments.len(),
full_audio_segments.len()
))?;
on_reset_segments()?;
segments.iter().cloned().try_for_each(&mut on_segment)?;
}
}
@ -327,10 +333,10 @@ fn should_prefer_full_audio(
let vad_end = last_end(vad_segments);
let full_end = last_end(full_audio_segments);
full_text_len > vad_text_len + vad_text_len / 5
|| full_audio_segments.len() > vad_segments.len() + 2
|| full_end > vad_end + 2.0
|| (total_seconds > 30.0 && full_end + 1.5 >= total_seconds && vad_end + 3.0 < total_seconds)
full_text_len > vad_text_len + vad_text_len * 3 / 5
|| full_audio_segments.len() > vad_segments.len() + 5
|| full_end > vad_end + 5.0
|| (total_seconds > 60.0 && full_end + 1.5 >= total_seconds && vad_end + 5.0 < total_seconds)
}
fn resolve_source_language<'a>(