Use whisper speech to text on an audio or video file regardless of codec, autodetect language

whisper-file

#!/bin/bash

# Usage: whisper-file FILE [LANGUAGE] [OPTIONS...]
# If LANGUAGE is empty, it is set to "auto"
# Any additional options are passed directly to ${WHISPER_BIN}

# General settings (paths) for whisper.cpp
# Note - this uses whisper.cpp, not official whisper. Get it at
# https://github.com/ggerganov/whisper.cpp
WHISPER_MODEL=/Users/test/whisper.cpp/models/ggml-large.bin
WHISPER_BIN=/Users/test/whisper.cpp/main

# File and language setup
FILE="${1}"
RECOGNITION_LANGUAGE="auto"
if [ -n "${2}" ]; then RECOGNITION_LANGUAGE="${2}"; fi

# Collect optional parameters
shift 2
EXTRA_PARAMS="$@"

WAV_RECODED_FILENAME="${FILE}-whisper.wav"
WAV_VTT_FILENAME="${WAV_RECODED_FILENAME}.vtt"

# Check if VTT already exists
if [ -e "${WAV_VTT_FILENAME}" ]; then
    echo "${WAV_VTT_FILENAME} already exists"
    exit 1
fi

# Check if WAV already exists
if [ -e "${WAV_RECODED_FILENAME}" ]; then
    echo "${WAV_RECODED_FILENAME} already exists, using it."
else
    ffmpeg -i "${FILE}" -vn -ar 16000 -ac 1 -c:a pcm_s16le "${WAV_RECODED_FILENAME}"
fi

# Run whisper.cpp with additional parameters if provided
${WHISPER_BIN} -m "${WHISPER_MODEL}" -f "${WAV_RECODED_FILENAME}" -otxt -ovtt -pp -l "${RECOGNITION_LANGUAGE}" ${EXTRA_PARAMS} && rm -f "${WAV_RECODED_FILENAME}"