A script to download an audio from a video from a streaming platform such as youtube and transcribe it to text using whisper.cpp

yt-whisper

#!/bin/bash

# Usage: yt-whisper URL [OUTPUT_FILENAME_TEMPLATE [LANGUAGE]]
# If OUTPUT_FILENAME_TEMPLATE is empty, output is yt-whisper-video
# 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
# You will have to adjust these
WHISPER_MODEL=/Users/test/whisper.cpp/models/ggml-large.bin
WHISPER_BIN=/Users/test/whisper.cpp/main
# Don't forget to install yt-dlp:
# brew install yt-dlp
YT_DOWNLOAD=yt-dlp

OUTPUT_BASENAME="yt-whisper-video"
if [ -n "${2}" ]; then OUTPUT_BASENAME="${2}"; fi

RECOGNITION_LANGUAGE="auto"
if [ -n "${3}" ]; then RECOGNITION_LANGUAGE="${3}"; fi

WAV_DOWNLOADED_FILENAME="${OUTPUT_BASENAME}-downloaded.wav"
WAV_RECODED_FILENAME="${OUTPUT_BASENAME}.wav"

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

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


${YT_DOWNLOAD} -x --audio-format wav -o "${WAV_DOWNLOADED_FILENAME}" "${1}"

if [ -not -f "${WAV_DOWNLOADED_FILENAME}" ]
    then
        echo "Download failed"
        exit 1
    fi

ffmpeg -i "${WAV_DOWNLOADED_FILENAME}" -ar 16000 -ac 1 -c:a pcm_s16le "${WAV_RECODED_FILENAME}" && rm -f "${WAV_DOWNLOADED_FILENAME}"

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