Transcribe all uni lections from a s3 bucket with whisper cpp

transcribe_lections.sh

#!/bin/bash

S3_BUCKET="hpc-lections"
WHISPER_CPP_PATH="whisper.cpp"
MODEL="large-v3"
LANG="ru"
AUDIO_EXTRACT=true

command -v aws >/dev/null 2>&1 || { echo >&2 "AWS CLI required."; exit 1; }
command -v ffmpeg >/dev/null 2>&1 && HAS_FFMPEG=true || HAS_FFMPEG=false

if [ "$AUDIO_EXTRACT" = true ] && [ "$HAS_FFMPEG" = false ]; then
    echo >&2 "FFmpeg required."; exit 1
fi

TMP_DIR=$(mktemp -d)
FILES=$(aws s3 ls "s3://${S3_BUCKET}" --recursive | awk '/\.mp4$/ {print $4}')

[ -z "$FILES" ] && { echo "No MP4 files found."; exit 0; }

for FILE in $FILES; do
    FILENAME=$(basename "$FILE")
    BASENAME="${FILENAME%.*}"

    LOCAL_MP4="$TMP_DIR/$FILENAME"
    LOCAL_AUDIO="$TMP_DIR/$BASENAME.wav"
    LOCAL_TRANSCRIPT="$TMP_DIR/$BASENAME.txt"

    aws s3 cp "s3://${S3_BUCKET}/${FILE}" "$LOCAL_MP4"

    if [ "$AUDIO_EXTRACT" = true ]; then
        ffmpeg -i "$LOCAL_MP4" -ar 16000 -ac 1 -c:a pcm_s16le "$LOCAL_AUDIO" -y
        INPUT_FILE="$LOCAL_AUDIO"
    else
        INPUT_FILE="$LOCAL_MP4"
    fi

    "$WHISPER_CPP_PATH/build/bin/whisper-cli" -m "$WHISPER_CPP_PATH/models/ggml-$MODEL.bin" -f "$INPUT_FILE" -l "$LANG" \
            -otxt -ml 1 2>&1 | tee "$LOCAL_TRANSCRIPT"

    S3_TRANSCRIPT_PATH="${FILE%.*}.txt"
    aws s3 cp "$LOCAL_TRANSCRIPT" "s3://${S3_BUCKET}/${S3_TRANSCRIPT_PATH}"

    rm -f "$LOCAL_MP4" "$LOCAL_AUDIO" "$LOCAL_TRANSCRIPT"
done

rm -rf "$TMP_DIR"