I am working on processing real-time audio from Amazon Connect by retrieving fragments from Kinesis Video Streams and saving them to S3. However, the MKV file I export to S3 is not playable and seems to be corrupted.

import boto3
import os
s3_bucket = 'BUCKET_NAME'

def lambda_handler(event, context):

    MediaStreams = event['Details']['ContactData']['MediaStreams']
    CustomerAudio = MediaStreams['Customer']['Audio']

    StartFragmentNumber = CustomerAudio['StartFragmentNumber']
    StreamARN = CustomerAudio['StreamARN']

    kvs_client = boto3.client("kinesisvideo", region_name="eu-central-1")
    endpoint = kvs_client.get_data_endpoint(
        StreamARN=StreamARN,
        APIName="GET_MEDIA"
    )["DataEndpoint"]

    media_client = boto3.client("kinesis-video-media", endpoint_url=endpoint, region_name="eu-central-1")
    response = media_client.get_media(
        StreamARN=StreamARN,
        StartSelector={
            "StartSelectorType": "FRAGMENT_NUMBER",
            "AfterFragmentNumber": StartFragmentNumber,
        }
    )

    process_audio(response["Payload"])

def process_audio(payload):
    s3_client = boto3.client('s3')
    s3_key = 'stream.mkv'
    temp_audio_path = '/tmp/stream.mkv'

    with open(temp_audio_path, 'wb') as f:
        f.write(payload.read())

    s3_client.upload_file(temp_audio_path, s3_bucket, s3_key)
    os.remove(temp_audio_path)
    return {
        'statusCode': 200,
        'body': f'Audio file saved to s3://{s3_bucket}/{s3_key}'
    }

The stream.mkv file saved in S3 is not playable. It seems corrupt, and standard media players cannot open it. I suspect that the data stream is not being correctly read or written.

How can I correctly parse and save these real-time media fragments?