chatgpttranscription

#!/bin/bash

# Function to show the usage of the script
show_usage() {
    echo "Usage: $(basename $0) [-m model] [-l language] [-p prompt] [-r response_format] [-o output_file] <audio_file>"
    echo "  -m  Model ID to use for transcription. Default is 'whisper-1'."
    echo "  -l  Language code of the input audio. Optional but improves accuracy."
    echo "  -p  Text to guide the model's style. Optional."
    echo "  -r  Response format of the transcript output. Options: text, vtt, json, srt, verbose_json. Default is 'text'."
    echo "  -o  Output file to write the response to. If not provided, the response will be printed to stdout."
    echo "  -t  Trim the output - remove any whitespace at beginning and end of the response."
    echo "  -h  Show this help message and exit."
    echo "  <audio_file> The audio file to transcribe."
    exit 1
}

# If the file exceeds 26200000 bytes: first downmix stereo to mono, then recode if still too large.
# In the extreme case: ffmpeg -i audio.mp3 -vn -map_metadata -1 -ac 1 -c:a libopus -b:a 12k -application voip audio.ogg

# Initialize default values
model="whisper-1"
response_format="text"
output_file=""
trim_output=0

# Parse options
while getopts ":m:l:p:r:o:th" opt; do
  case $opt in
    m) model="$OPTARG"
    ;;
    l) language="$OPTARG"
    ;;
    p) prompt="$OPTARG"
    ;;
    r) response_format="$OPTARG"
    ;;
    o) output_file="$OPTARG"
    ;;
    h) show_usage
    ;;
    \?) echo "Invalid option -$OPTARG" >&2
        show_usage
    ;;
    :) echo "Option -$OPTARG requires an argument." >&2
       show_usage
    ;;
  esac
done

# Remove the options from the positional parameters
shift $((OPTIND -1))

# Check for the audio file argument
if [ "$#" -ne 1 ]; then
    echo "Error: You must provide an audio file to transcribe."
    show_usage
fi

audio_file=$1

# Verify that the audio file exists
if [ ! -f "$audio_file" ]; then
    echo "Error: The file $audio_file does not exist."
    exit 1
fi

# downmix to mono if larger than 26200000 and file is stereo, recode if still too large
if [ $(wc -c < "$audio_file") -gt 26200000 ]; then
  #tmpdir is $TMPDIR or /tmp
  tmpdir=$TMPDIR
  [ -z "$tmpdir" ] && tmpdir=/tmp
  trap 'rm -f "$tmpfile" "$tmpfile2"' EXIT

  # check number of audio channels
  channels=$(ffprobe -v error -select_streams a:0 -show_entries stream=channels -of csv=p=0 "$audio_file" 2>/dev/null)
  if [ "$channels" -gt 1 ] 2>/dev/null; then
    tmpfile=$(mktemp -u "$tmpdir/tmp.XXXXXXXXXX").mp3
    echo "Downmixing stereo to mono..."
    ffmpeg -i "$audio_file" -vn -map_metadata -1 -ac 1 -c:a libmp3lame "$tmpfile"
    audio_file=$tmpfile
    ls -l "$audio_file"
  fi

  # recode if still too large
  if [ $(wc -c < "$audio_file") -gt 26200000 ]; then
    tmpfile2=$(mktemp -u "$tmpdir/tmp.XXXXXXXXXX").mp3
    echo "Recoding to reduce file size..."
    # if lame is in the path, use it to resample and reduce the bitrate
    if command -v lame &> /dev/null; then
      lame -h -a -S -t --resample 12 --abr 16 -m m "$audio_file" "$tmpfile2"
    else
      ffmpeg -i "$audio_file" -vn -map_metadata -1 -ac 1 -c:a libopus -b:a 12k -application voip "$tmpfile2"
    fi
    audio_file=$tmpfile2
    ls -l "$audio_file"
  fi
fi

# Read the OpenAI API key
if [ -z "$OPENAI_API_KEY" ]; then
    if [ -f "$HOME/.openai-api-key.txt" ]; then
        OPENAI_API_KEY=$(cat "$HOME/.openai-api-key.txt")
    else
        echo "Error: OPENAI_API_KEY is not set and no config file found at $HOME/.openai-api-key.txt."
        exit 1
    fi
fi

# Build the curl command
curl_command="curl -X POST 'https://api.openai.com/v1/audio/transcriptions' -s -S"
curl_command+=" -H 'Authorization: Bearer $OPENAI_API_KEY'"
curl_command+=" -H 'Content-Type: multipart/form-data'"
curl_command+=" -F 'file=@$audio_file'"
curl_command+=" -F 'model=$model'"

# Add optional parameters if provided
[ -n "$language" ] && curl_command+=" -F 'language=$language'"
[ -n "$prompt" ] && curl_command+=" -F 'prompt=$prompt'"
curl_command+=" -F 'response_format=$response_format'"

# Execute the curl command and handle the response
if [ -n "$output_file" ]; then
    eval "$curl_command" -o "$output_file"
    echo "Response written to $output_file"
elif [ $trim_output -eq 1 ]; then
    eval "$curl_command" | sed -e 's/^[[:space:]]*//' -e 's/[[:space:]]*$//'
else
    eval "$curl_command"
fi

exit 0