| import wespeakerruntime as wespeaker | |
| import os | |
| import numpy as np | |
| from tqdm import tqdm | |
| inference = wespeaker.Inference(lang='chs') | |
| wave_path = '/home/admin/yuanxin/3.tempData/FastSpeech2/raw_data/AISHELL3-Pro' | |
| def main(): | |
| with open('speaker_embedding.txt', 'w', encoding='utf-8') as f: | |
| speakers_path = os.listdir(wave_path) | |
| for speaker in tqdm(speakers_path): | |
| if speaker in ['cicely', 'xiaoyin']: | |
| continue | |
| s_list = os.listdir(os.path.join(wave_path, speaker)) | |
| temp = np.zeros((1, 256)) | |
| count = 0 | |
| for s in tqdm(s_list): | |
| if '.wav' in s: | |
| count += 1 | |
| temp += inference.extract_embedding_wav(os.path.join(wave_path, speaker, s)) | |
| else: | |
| continue | |
| f.write(speaker + "|" + str((temp / count).tolist()[0]) + '\n') | |
| if __name__ == "__main__": | |
| main() | |