auto-transcribe

Batch transcribe audio files using whisper-cpp with watch mode.

Prerequisites

• Node.js >= 18
• whisper-cpp (provides whisper-cli)

brew install whisper-cpp

Install

npm install -g auto-transcribe

Usage

auto-transcribe [command] [options] [input]

Commands

| Command | Description | | ------------ | ---------------------------------------- | | transcribe | Transcribe audio file(s) (default) | | watch | Watch directory and transcribe new files |

Options

| Option | Description | | ----------------------- | ------------------------------------------------------------ | | -m, --model <path> | Model file (.bin). Auto-detects from common locations | | -l, --language <lang> | Spoken language or auto (default: auto) | | -o, --output <dir> | Output directory (default: ./outputs) | | -f, --format <fmts> | Comma-separated: txt,ts,vtt,srt,lrc,csv,json,json-full,words | | --model-dir <dir> | Directory containing .bin model files | | --no-date-dir | Omit date subdirectory in output path | | -h, --help | Show help | | --version | Show version |

Examples

# Transcribe a single file
auto-transcribe recording.mp3

# Transcribe all audio files in a directory
auto-transcribe ./audio-files/

# Watch a directory for new audio files
auto-transcribe watch ./incoming/

# Specify language and output formats
auto-transcribe -l fr -f txt,srt interview.wav

Supported audio formats

.flac, .mp3, .ogg, .wav, .m4a, .aac, .opus

Output formats

txt, ts, vtt, srt, lrc, csv, json, json-full, words (default: txt,json,vtt)

Download models from huggingface.co/ggerganov/whisper.cpp:

mkdir -p models
curl -L -o models/ggml-base.en.bin \
  https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin