faster-whisper

Overview

Local speech-to-text using faster-whisper, a CTranslate2 reimplementation of OpenAI’s Whisper, for fast and accurate transcription with GPU acceleration.

Key Features

• Transcribe audio/video files
• Generate subtitles (SRT, VTT, ASS, LRC, TTML)
• Identify speakers (diarization labels)
• Transcribe from URLs (YouTube links and direct audio URLs)
• Batch process files (glob patterns, directories, skip-existing support)
• Convert speech to text locally (no API costs, works offline)
• Translate to English
• Do multilingual transcription (supports 99+ languages with auto-detection)
• Transcribe a batch of files in different languages
• Transcribe multilingual audio
• Transcribe audio with specific terms
• Preprocess noisy audio (before transcription)
• Stream output
• Clip time ranges
• Search the transcript
• Detect chapters
• Export speaker audio
• Spreadsheet output

How It Works

Use the faster-whisper skill to transcribe audio/video files, generate subtitles, and more. The skill uses the faster-whisper model, which runs 4-6x faster than OpenAI Whisper with identical accuracy. With GPU acceleration, expect ~20x realtime transcription.

Use Cases

• Transcribe a meeting or interview
• Generate subtitles for a YouTube video
• Identify speakers in a podcast
• Transcribe a batch of files in different languages
• Transcribe multilingual audio
• Preprocess noisy audio before transcription
• Stream output for real-time transcription
• Clip time ranges for specific sections
• Search the transcript for specific terms
• Detect chapters for a table of contents
• Export speaker audio for separate WAV files
• Spreadsheet output for CSV or spreadsheet format