重建效果是否符合预期

[zzchust]

测试了一条ucg的音频， 使用75token-large模型重建音频，效果很差，可以帮忙看看不？
测试音频-原始音频+重建音频.zip

[zzchust]

`from encoder.utils import convert_audio
import torchaudio
import torch
from decoder.pretrained import WavTokenizer

device=torch.device('cuda')

config_path = "configs/wavtokenizer_smalldata_frame75_3s_nq1_code4096_dim512_kmeans200_attn.yaml"
model_path = "WavTokenizer-large-speech-75token/wavtokenizer_large_speech_320_24k.ckpt"
audio_path = "souce_audio.mp3"

audio_outpath = "reconstruct_audio.wav"

wavtokenizer = WavTokenizer.from_pretrained0802(config_path, model_path)
wavtokenizer = wavtokenizer.to(device)

torchaudio.set_audio_backend('soundfile')

wav, sr = torchaudio.load(audio_path)
wav = convert_audio(wav, sr, 24000, 1)
bandwidth_id = torch.tensor([0])
wav=wav.to(device)
features,discrete_code= wavtokenizer.encode_infer(wav, bandwidth_id=bandwidth_id)
audio_out = wavtokenizer.decode(features.cuda(), bandwidth_id=bandwidth_id.cuda())
torchaudio.save(audio_outpath, audio_out.cpu(), sample_rate=24000, encoding='PCM_S', bits_per_sample=16)
`

[jishengpeng]

测试了一条ucg的音频， 使用75token-large模型重建音频，效果很差，可以帮忙看看不？ 测试音频-原始音频+重建音频.zip

large版本中的训练中文数据也很少，得用干净中文重新训练，issue区里有很多人已经试过了。

[zzchust]

好的，感谢回复！