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]
# If LANGAUGE is empty, it is set to "auto"

# 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

RECOGNITION_LANGUAGE="auto"
if [ -n "${2}" ]; then RECOGNITION_LANGUAGE="${2}"; fi

WAV_RECODED_FILENAME="${1}-whisper.wav"

WAV_VTT_FILENAME="${WAV_RECODED_FILENAME}.vtt"

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

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

${WHISPER_BIN} -m ${WHISPER_MODEL} -f "${WAV_RECODED_FILENAME}" -otxt -ovtt -pp -l "${RECOGNITION_LANGUAGE}" && rm -f "${WAV_RECODED_FILENAME}"