2023-05-29 17:52:23 +02:00
|
|
|
import soundfile
|
|
|
|
|
from infer_pack.onnx_inference import OnnxRVC
|
|
|
|
|
|
|
|
|
|
hop_size = 512
|
2023-05-30 09:22:53 +02:00
|
|
|
sampling_rate = 40000 # 采样率
|
|
|
|
|
f0_up_key = 0 # 升降调
|
|
|
|
|
sid = 0 # 角色ID
|
|
|
|
|
f0_method = "dio" # F0提取算法
|
|
|
|
|
model_path = "ShirohaRVC.onnx" # 模型的完整路径
|
|
|
|
|
vec_name = "vec-256-layer-9" # 内部自动补齐为 f"pretrained/{vec_name}.onnx" 需要onnx的vec模型
|
|
|
|
|
wav_path = "123.wav" # 输入路径或ByteIO实例
|
|
|
|
|
out_path = "out.wav" # 输出路径或ByteIO实例
|
2023-05-29 17:52:23 +02:00
|
|
|
|
2023-05-30 09:22:53 +02:00
|
|
|
model = OnnxRVC(
|
|
|
|
|
model_path, vec_path=vec_name, sr=sampling_rate, hop_size=hop_size, device="cuda"
|
|
|
|
|
)
|
2023-05-29 17:52:23 +02:00
|
|
|
|
|
|
|
|
audio = model.inference(wav_path, sid, f0_method=f0_method, f0_up_key=f0_up_key)
|
|
|
|
|
|
2023-05-30 09:22:53 +02:00
|
|
|
soundfile.write(out_path, audio, sampling_rate)
|
