Using tensorflow to realize voiceprint recognition

Preface

This paper introduces the use of tensorflow Implement a simple voiceprint recognition model , First, you need to be familiar with audio classification , If you don't know, you can check my last article - be based on tensorflow Realize sound classification , On this basis , We train a voiceprint recognition model , Through this model, we can identify who the speaker is , It can be applied to some audio verification projects . The difference is that this project uses ArcFace loss,ArcFace loss：Additive Angular Margin Loss（ Additive angular interval loss function ）, Normalize the feature vector and weight , Yes θ Plus angular spacing , The influence of angle interval on angle is more direct than cosine interval .
Usage environment
python3.8
tensorflow2.3.0

Model download

https://download.csdn.net/download/qq_33200967/20368421

install

1.pip install tensorflow==2.3.0 -i https://mirrors.aliyun.com/pypi/simple/
2.pip install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/

Create data

1. This niuma uses a Chinese speech corpus , This dataset has 3342 Personal voice data , Yes 1130000+ Voice data . If the reader has other better data sets , Can be mixed together , Use python Tool module of aukit Processing audio 、 Noise reduction and mute removal
2. First create a data list ,** The format of the data list is < Voice file path / Voice tag >, Creating this list is mainly It is convenient to read later , It is also convenient to read other voice data sets , Voice classification tags are the unique identification of different people id, Different voice data sets , You can write the corresponding function to generate the data list , Write these data sets in the data list
3. stay create_data.py Write down the code in , Because Chinese speech corpus This data set is mp3 Format , Ben niuma found that the reading speed of this format is very slow , So I put all mp3 Format of audio conversion to wav Format , After creating the data list , Some data may be wrong , So we need to check , Delete the wrong data . Execute the following procedure to complete data preparation

import json
import os

from pydub import AudioSegment
from tqdm import tqdm

from utils.reader import load_audio


#  Generate data list 
def get_data_list(infodata_path, list_path, zhvoice_path):
    with open(infodata_path, 'r', encoding='utf-8') as f:
        lines = f.readlines()

    f_train = open(os.path.join(list_path, 'train_list.txt'), 'w')
    f_test = open(os.path.join(list_path, 'test_list.txt'), 'w')

    sound_sum = 0
    speakers = []
    speakers_dict = {
    }
    for line in tqdm(lines):
        line = json.loads(line.replace('\n', ''))
        duration_ms = line['duration_ms']
        if duration_ms < 1300:
            continue
        speaker = line['speaker']
        if speaker not in speakers:
            speakers_dict[speaker] = len(speakers)
            speakers.append(speaker)
        label = speakers_dict[speaker]
        sound_path = os.path.join(zhvoice_path, line['index'])
        save_path = "%s.wav" % sound_path[:-4]
        if not os.path.exists(save_path):
            try:
                wav = AudioSegment.from_mp3(sound_path)
                wav.export(save_path, format="wav")
                os.remove(sound_path)
            except Exception as e:
                print(' Data error ：%s,  Information ：%s' % (sound_path, e))
                continue
        if sound_sum % 200 == 0:
            f_test.write('%s\t%d\n' % (save_path.replace('\\', '/'), label))
        else:
            f_train.write('%s\t%d\n' % (save_path.replace('\\', '/'), label))
        sound_sum += 1

    f_test.close()
    f_train.close()


#  Delete error audio 
def remove_error_audio(data_list_path):
    with open(data_list_path, 'r', encoding='utf-8') as f:
        lines = f.readlines()
    lines1 = []
    for line in tqdm(lines):
        audio_path, _ = line.split('\t')
        try:
            spec_mag = load_audio(audio_path)
            lines1.append(line)
        except Exception as e:
            print(audio_path)
            print(e)
    with open(data_list_path, 'w', encoding='utf-8') as f:
        for line in lines1:
            f.write(line)


if __name__ == '__main__':
    get_data_list('dataset/zhvoice/text/infodata.json', 'dataset', 'dataset/zhvoice')
    remove_error_audio('dataset/train_list.txt')
    remove_error_audio('dataset/test_list.txt')

After executing the above code , The following data formats will be generated ,, If you want to customize data , Refer to the following data list , The front is the relative path of audio , Behind it is the label of the speaker , for example
dataset/zhvoice/zhmagicdata/5_895/5_895_20170614203758.wav 3238

data fetch