Dictate to OpenAI's Whisper API from the cmdline

dict.sh

#!/bin/bash

# @author: github.com/jckw
# @dependencies:
# - sox (brew install sox)

# Define constants
TMP_FILE="/tmp/openai_recording.mp3"
OPENAI_KEY="YOUR_OPENAI_API_KEY"

# Function to handle Ctrl+C interruption and stop recording
interrupt_recording() {
    display_message "Stopping recording..."
    # Kill the sox recording process
    kill $SOX_PID
}

# Function to display formatted messages
display_message() {
    echo -e "\n----------------------------------"
    echo "$1"
    echo -e "----------------------------------\n"
}

# Set the trap for interruption
trap interrupt_recording SIGINT

# Start the audio recording
display_message "Recording... Press Ctrl+C to stop."
rec $TMP_FILE &
# Get the process ID of sox for interruption handling
SOX_PID=$!

# Wait for the recording to finish
wait $SOX_PID

# A short delay to ensure the file is completely written
sleep 1

# Send the recording to OpenAI and extract the 'text' field
display_message "Sending the recording to OpenAI..."
RESPONSE_TEXT=$(curl --request POST \
    --url https://api.openai.com/v1/audio/transcriptions \
    --header "Authorization: Bearer $OPENAI_KEY" \
    --header 'Content-Type: multipart/form-data' \
    --form "file=@$TMP_FILE" \
    --form model=whisper-1 \
    --form response_format=text
)

# Check if API request was successful
if [ $? -ne 0 ]; then
    echo "Failed to send recording or parse response."
    exit 1
fi

# Copy the transcription to the clipboard (assuming pbcopy is available on MacOS)
echo $RESPONSE_TEXT | pbcopy
display_message "Transcription copied to clipboard!"
echo $RESPONSE_TEXT

# Cleanup the temporary recording file
rm $TMP_FILE