From 07058e17c34e2b694c984381dfda4f2c7a14e629 Mon Sep 17 00:00:00 2001 From: Chocobozzz Date: Wed, 3 Jul 2024 17:08:32 +0200 Subject: [PATCH] Better whisper-ctranslate vad filter option --- .../src/whisper/transcriber/ctranslate2-transcriber.ts | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/packages/transcription/src/whisper/transcriber/ctranslate2-transcriber.ts b/packages/transcription/src/whisper/transcriber/ctranslate2-transcriber.ts index de0a9c508..6a55b10e3 100644 --- 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',