Whisper.ccp transcription helper

README.md

Whisper.cpp transcribe script

This is a helper script to make the process of transcribing with https://github.com/ggerganov/whisper.cpp model easier.

Installation

1. Clone the https://github.com/ggerganov/whisper.cpp repo
2. Follow the setup instructions
3. Paste the transcribe.sh script from this gist in the root dir of whisper.cpp project

transcribe.sh

#!/bin/bash

FILE_TO_TRANSCRIBE="$1"
FILE_TO_TRANSCRIBE_WAV="${1%.*}.wav"

LANGUAGE="${2:-en}"

HELP_TEXT=$(cat <<'EOF'
Usage: transcribe.sh <file> <lang>

Transcribe the given video or audio file.
Positional arguments:
  file - video or audio file
  lang - Specify the language of the audio file.
    Defaults to English (en). Use 'auto' to identify the lang.
optional arguments:
  -s, --skip-convert  Skip the conversion to 16-bit WAV file
  -h, --help  Show this help message and exit
EOF
)

if [ "$FILE_TO_TRANSCRIBE" = "" ]; then
  echo "$HELP_TEXT"
  exit 1
fi

SHOULD_CONVERT="1"
for flag in "$@"; do
  case "$flag" in
    -s|--skip-convert)
      SHOULD_CONVERT="0"
      ;;
    -h|--help)
      echo "$HELP_TEXT"
      exit 0
      ;;
  esac
done


if [ "$SHOULD_CONVERT" = "1" ]; then
  # Convert to 16-bit WAV
  ffmpeg -i "$FILE_TO_TRANSCRIBE" -ar 16000 -ac 1 -c:a pcm_s16le "$FILE_TO_TRANSCRIBE_WAV"
fi


if [ -f "$FILE_TO_TRANSCRIBE_WAV" ]; then
  ./main --model models/ggml-large.bin --file "$FILE_TO_TRANSCRIBE_WAV" --language "$LANGUAGE" --output-srt
else
  ./main --model models/ggml-large.bin --file "$FILE_TO_TRANSCRIBE" --language "$LANGUAGE" --output-srt
fi