Whisper Speech to Text Shellscript

speech2text-whisper.sh

#!/bin/bash
# Speech to text with Whisper model using OpenAI API
# Assumes that the OpenAI API key is set as an environment variable
# Usage: ./speech2text-whisper.sh input_audio_file

# Check if an input argument is provided and validate the file extension
if [ -z "$1" ]; then
  echo "Error: No input file provided. Please specify an audio file (MP3, WAV, etc.) as the first argument."
  exit 1
fi

if [[ ! "$1" =~ \.(mp3|wav|ogg|flac)$ ]]; then
  echo "Error: The input file must be an audio file (MP3, WAV, OGG, FLAC)."
  exit 1
fi

infile_path=$1


# Check if the file size exceeds 25MB
max_size_bytes=$((25 * 1024 * 1024))
file_size_bytes=$(du -b "$infile_path" | cut -f1)

if [ $file_size_bytes -gt $max_size_bytes ]; then
  echo "Error: The input file size exceeds 25MB. Please use a file that is 25MB or smaller."
  exit 1
fi


# Set your OpenAI API key
#$OPENAI_API_KEY="sk-..."

# Set the Whisper model to use
WHISPER_MODEL="whisper-1"

# Set the output file name
outfile_path="${infile_path%.*}_transcription.txt"

# Create a temporary response file
now=$(date +"%Y%m%d_%H%M%S")
temp_response_file=response_$now.txt
RESPONSE_FORMAT="text"

# Submit the request to the OpenAI API and store the response in a temporary file
curl -s https://api.openai.com/v1/audio/transcriptions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: multipart/form-data" \
  -F "model=$WHISPER_MODEL" \
  -F "response_format=$RESPONSE_FORMAT" \
  -F "temperature=0" \
  -F "language=en" \
  -F "file=@$infile_path" > $outfile_path

if [ -n "$outfile_path" ]; then
  echo "Transcription successful. Saved to $outfile_path..."
  
else
  echo "Error: Could not extract transcription text from API response"
  
fi