Better whisper-ctranslate vad filter option

Chocobozzz · Jul 3, 2024 · 07058e1 · 07058e1
1 parent efa5835
commit 07058e1
Showing 1 changed file with 4 additions and 0 deletions.
diff --git a/packages/transcription/src/whisper/transcriber/ctranslate2-transcriber.ts b/packages/transcription/src/whisper/transcriber/ctranslate2-transcriber.ts
@@ -37,6 +37,10 @@ export class Ctranslate2Transcriber extends OpenaiTranscriber {
       'True',
       '--vad_filter',
       'true',
+      // Better precision with 5s of audio
+      // We mainly use vad_filter to improve language detection (first 30 seconds of the video, so no voice is problematic)
+      '--vad_min_silence_duration_ms',
+      '5000',
       '--output_format',
       'all',
       '--output_dir',