import torch,pdb
import numpy as np
import soundfile as sf
from models import SynthesizerTrnNoF0256
from scipy.io import wavfile
from fairseq import checkpoint_utils
import torch.nn.functional as F

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model_path = "https://ibm.box.com/s/z1wgl1stco8ffooyatzdwsqn2psd9lrr"#checkpoint_best_legacy_500.pt
print("load model(s) from {}".format(model_path))
models, saved_cfg, task = checkpoint_utils.load_model_ensemble_and_task(
    [model_path],
    suffix="",
)
model = models[0]
model = model.to(device)
model = model.half()
model.eval()

net_g = SynthesizerTrnNoF0256(513,40,192,192,768,2,6,3,0.1,"1", [3,7,11],[[1,3,5], [1,3,5], [1,3,5]],[10,4,2,2,2],512,[16,16,4,4,4],0)
weights=torch.load("trump.pt")
net_g.load_state_dict(weights,strict=True)
net_g.eval().to(device)
net_g.half()


wav_path="/bili-coeus/liujing04/vits_ch/测试素材/trump/wavs16k/云希_特朗普.wav"
wav, sr = sf.read(wav_path)
assert sr == 16000
feats = torch.from_numpy(wav).float()
if feats.dim() == 2:  # double channels
    feats = feats.mean(-1)
assert feats.dim() == 1, feats.dim()
feats = feats.view(1, -1)
padding_mask = torch.BoolTensor(feats.shape).fill_(False)
inputs = {
    "source": feats.half().to(device),
    "padding_mask": padding_mask.to(device),
    "output_layer": 9,  # layer 9
}
with torch.no_grad():
    logits = model.extract_features(**inputs)
    feats = model.final_proj(logits[0])
feats=F.interpolate(feats.permute(0,2,1),scale_factor=2).permute(0,2,1)
p_len = min(feats.shape[1],10000)#太大了爆显存
feats = feats[:,:p_len, :]
p_len = torch.LongTensor([p_len]).to(device)
with torch.no_grad():
    audio = net_g.infer(feats, p_len)[0][0, 0].data.cpu().float().numpy()

wavfile.write("trump_co256nof0_63k_test.wav", 32000, audio)