db/make_audio.py

import io

from pymilvus import Collection
import asyncio
from bark import SAMPLE_RATE, generate_audio, preload_models
from scipy.io.wavfile import write
from db_connect import connect

# for audio storage
from audio_db.is3.is3 import StagedObject

connect()
preload_models()


async def make_audio():
    response = Collection("Response")
    audio_response = Collection("AudioResponse")

    data = []

    response_iterator = response.query_iterator(batch_size=1, output_fields=['text', 'embeddings'])
    ids_to_delete = []
    current_iteration = 0
    batch_size = 5

    while True:
        res = response_iterator.next()
        if len(res) == 0:
            print("query iteration finished, close")
            # close the iterator
            response_iterator.close()
            break

        # generate audio
        audio_array = generate_audio(res[0]['text'], history_prompt="en_speaker_3")
        bytes_io = io.BytesIO()
        write(bytes_io, SAMPLE_RATE, audio_array)
        audio_bytes = bytes_io.read()

        # store the audio
        obj = StagedObject(obj=audio_bytes, name='audio')
        uploaded_object = await obj.upload()

        # save the audio record to AudioResponse
        data.append([res[0]['text'], uploaded_object.obj_id, res[0]['embeddings']])
        ids_to_delete.append(res[0]['id'])

        current_iteration += 1

        if current_iteration % batch_size == 0:
            # insert the audio
            audio_response.insert(list(zip(*data)))
            audio_response.flush()

            # delete text to generate audio
            response.delete(expr=f"id in {str(ids_to_delete)}")
            ids_to_delete = []

    response_iterator.close()

if __name__ == '__main__':
    loop = asyncio.new_event_loop()
    loop.run_until_complete(make_audio())