How to transcribe audios bigger than 25MB with Whisper AI

index.py

import openai
import mimetypes
from pydub import AudioSegment
import os

load_dotenv()
openai.api_key = os.getenv('OPEN_AI_KEY')


app = Flask(__name__)
logging.basicConfig(level=logging.DEBUG, format='%(asctime)s %(levelname)s: %(message)s')


@app.route('/split-audio', methods=['POST'])
def split_audio():
    try:
        audio_file = request.files['audio']
        file_extension = mimetypes.guess_extension(audio_file.content_type)[1:]
        file_format = mimetypes.guess_extension(file_extension)

        audio = AudioSegment.from_file(audio_file, format=file_format)

        segment_length = 10 * 60  # Start with 10 minutes
        max_segment_size = 24 * 1024 * 1024  # 24MB in bytes
        prompt = ""
        full_text = ""

        segment_filename = os.path.splitext(audio_file.filename)[0]
        duration = audio.duration_seconds
        number_of_segments = int(duration / segment_length)

        logging.info('Splitting audio into segments')

        if number_of_segments == 0:
            number_of_segments = 1

        # Split the audio into segments and transcribe each segment
        i = 0
        segment_start = 0
        segment_end = segment_length * 1000

        while segment_start < duration * 1000:
            logging.info('Segment ' + str(i+1) + ' processing')

            sound_export = audio[segment_start:segment_end]
            export_format = file_extension
            exported_file = '/tmp/' + segment_filename + '-' + str(i+1) + '.' + export_format
            sound_export.export(exported_file, format=export_format)

            # Check the size of the exported file
            while os.path.getsize(exported_file) > max_segment_size:
                # If the file is larger than 24MB, split it into smaller segments
                segment_length /= 2  # Halve the segment length
                segment_end = segment_start + segment_length * 1000
                sound_export = audio[segment_start:segment_end]
                sound_export.export(exported_file, format=export_format)

            logging.info('Segment ' + str(i+1) + ' exported')

            with open(exported_file, "rb") as f:
                data = openai.Audio.transcribe("whisper-1", f, prompt=prompt)

            logging.info('Segment ' + str(i+1) + ' transcribing')

            prompt += data.text
            full_text += data.text

            logging.info('Segment ' + str(i+1) + ' transcribed')

            # Update segment_start and segment_end for the next loop iteration
            segment_start = segment_end
            segment_end += segment_length * 1000
            i += 1

        logging.info('Audio transcribed')

        return jsonify({'full_text': full_text}), 200

    except Exception as e:
        error_message = 'Error occurred: ' + str(e)
        logging.error(error_message)
        response = jsonify({'error': str(e)})
        response.status_code = 400  # Set the status code to indicate a client error
        return response

check my blog post for more in depth explanation of the code, any comments please let me know!