main_pipeline.py

import argparse
import librosa
import torch
import os

from scipy.io.wavfile import write
from tqdm import tqdm

from utils.denoise_pipeline import denoise
from utils.diarization_pipeline import DiarizationPipeline

import numpy as np

import pandas as pd
import soundfile as sf


class CleaningPipeline:
    def __init__(self, device):
        """
        Cleaning audio pipeline. Contains:
            - denoising
            - diarization
        """

        self.device = device
        self.denoiser = denoise
        self.diarization = DiarizationPipeline(device)

    def __call__(self, input_audio_path: str):
        denoised_audio_path = self.denoiser(input_audio_path, self.device)
        result_diarization = self.diarization(denoised_audio_path)

        if result_diarization != {}:
            output_diar_audio_paths = result_diarization['output_diar_audio_paths']
            count_speakers = result_diarization['count_speakers']
            return [denoised_audio_path] + output_diar_audio_paths + [None] * (19 - count_speakers)

        else:
            return []


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--audio-path', default='dialog.mp3', help='Path to audio')
    parser.add_argument('--device', default='cpu', help='Path to audio')
    parser.add_argument('--out-folder-path', default='out', help='Path to result folder')
    opt = parser.parse_args()
    cleaning_pipeline = CleaningPipeline('cuda:0')
    cleaning_pipeline(input_audio_path=opt.audio_path)
    # for _ in tqdm(range(10)):
    #     main_pipeline(audio_path=opt.audio_path, device=opt.device)