#!/bin/zsh
export LD_LIBRARY_PATH="$(find /home/sacha/vendor/whisperx/.venv/lib/python3.12/site-packages/nvidia -maxdepth 2 -name lib -type d | tr '\n' ':')${LD_LIBRARY_PATH}"
WHISPER_ARGS=(${(z)WHISPER_FLAGS})
SPEAKERS=${SPEAKERS:-2}
MAX_LINE_WIDTH="${MAX_LINE_WIDTH:-50}"
MODEL="${MODEL:-large-v3}"
DEVICE=cpu
for FILE in "$@"; do
    text="${FILE%.*}.txt"
    if [  -f "$text" ]; then
       echo "Skipping $FILE as it's already been transcribed."
    else
				ffmpeg -y -i "$FILE" -map 0:a:1 -ac 1 -c:a copy "$(basename ${FILE%.*})-sacha.aac"
    		~/vendor/whisperx/.venv/bin/whisperx --model "$MODEL" --device $DEVICE --language fr --align_model WAV2VEC2_ASR_LARGE_LV60K_960H --compute_type int8 --print_progress True --segment_resolution chunk --language fr "$(basename ${FILE%.*})-sacha.aac" $WHISPER_FLAGS
				cp "$(basename ${FILE%.*})-sacha*" ~/proj/french/conversation/sacha-only
    		~/vendor/whisperx/.venv/bin/whisperx --model "$MODEL" --diarize --hf_token $HUGGING_FACE_API_KEY  --device $DEVICE --language fr --align_model WAV2VEC2_ASR_LARGE_LV60K_960H --compute_type int8 --print_progress True --max_line_width $MAX_LINE_WIDTH --segment_resolution chunk --max_line_count 1 --language fr "$FILE" $WHISPER_FLAGS --max-speakers $SPEAKERS
    		# ~/vendor/whisperx/.venv/bin/whisperx --model "$MODEL" --device $DEVICE --language fr --align_model WAV2VEC2_ASR_LARGE_LV60K_960H --compute_type int8 --print_progress True --max_line_width $MAX_LINE_WIDTH --segment_resolution chunk --max_line_count 1 --language fr "$FILE" $WHISPER_FLAGS

		    rm -f "${FILE%.*}.srt"
				cp "${FILE%.*}.json" "${FILE%.*}.txt" ~/proj/french/inbox
    fi
done
