BigVGAN模型迁移和优化指导

# 增加执行权限
chmod +x ./Ascend-cann-toolkit_{version}_linux-{arch}.run
chmod +x ./Ascend-cann-kernels-{soc}_{version}_linux.run

# 校验安装包
./Ascend-cann-toolkit_{version}_linux-{arch}.run --check
./Ascend-cann-kernels-{soc}_{version}_linux.run --check

# 执行安装
./Ascend-cann-toolkit_{version}_linux-{arch}.run --install
./Ascend-cann-kernels-{soc}_{version}_linux.run --install

# 设置环境变量
source /usr/local/Ascend/ascend-toolkit/set_env.sh

git clone https://github.com/NVIDIA/BigVGAN
cd BigVGAN
pip install -r requirements.txt

device = 'npu'
import torch
import bigvgan
import librosa
from meldataset import get_mel_spectrogram
import torch_npu
from torch_npu.contrib import transfer_to_npu
from zhl_bigvgan import BigVGAN

# NPU配置
torch_npu.npu.config.allow_internal_format=False
torch_npu.npu.set_compile_mode(jit_compile=False)

# 加载预训练模型
model = bigvgan.BigVGAN.from_pretrained('nvidia/bigvgan_v2_24khz_100band_256x', use_cuda_kernel=False)

# 模型准备
model.remove_weight_norm()
model = model.eval().to(device)

# 音频处理流程
wav_path = '/path/to/your/audio.wav'
wav, sr = librosa.load(wav_path, sr=model.h.sampling_rate, mono=True)
wav = torch.FloatTensor(wav).unsqueeze(0)

# 生成梅尔频谱
mel = get_mel_spectrogram(wav, model.h).to(device)

# 波形生成
with torch.inference_mode():
    wav_gen = model(mel)
    wav_gen_float = wav_gen.squeeze(0).cpu()

# 格式转换
wav_gen_int16 = (wav_gen_float * 32767.0).numpy().astype('int16')