app.py

import gradio as gr

import logging
numba_logger = logging.getLogger('numba')
numba_logger.setLevel(logging.WARNING)
import torch,pdb
import numpy as np
from models import SynthesizerTrnNoF0256
from fairseq import checkpoint_utils
import torch.nn.functional as F
import librosa

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model_path = "checkpoint_best_legacy_500.pt"#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.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", map_location=torch.device('cpu'))
net_g.load_state_dict(weights,strict=True)
net_g.eval().to(device)


def vc_fn( input_audio):
    if input_audio is None:
        return "You need to upload an audio", None
    sampling_rate, audio = input_audio
    duration = audio.shape[0] / sampling_rate
    if duration > 45:
        return "请上传小于45s的音频，需要转换长音频请使用colab", None
    audio = (audio / np.iinfo(audio.dtype).max).astype(np.float32)
    if len(audio.shape) > 1:
        audio = librosa.to_mono(audio.transpose(1, 0))
    if sampling_rate != 16000:
        audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000)

    print(audio.shape)

    feats = torch.from_numpy(audio).float()
    assert feats.dim() == 1, feats.dim()
    feats = feats.view(1, -1)
    padding_mask = torch.BoolTensor(feats.shape).fill_(False)
    inputs = {
        "source": feats.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()

    return "Success", (32000, audio)


app = gr.Blocks()
with app:
    with gr.Tabs():
        with gr.TabItem("Basic"):
            gr.Markdown(value="""""")
            vc_input3 = gr.Audio(label="上传音频（长度小于45秒）")
            vc_submit = gr.Button("转换", variant="primary")
            vc_output1 = gr.Textbox(label="Output Message")
            vc_output2 = gr.Audio(label="Output Audio")
        vc_submit.click(vc_fn, [ vc_input3], [vc_output1, vc_output2])

    app.launch()