|
import io |
|
import os |
|
|
|
|
|
import gradio as gr |
|
import librosa |
|
import numpy as np |
|
import soundfile |
|
from inference.infer_tool import Svc |
|
import logging |
|
|
|
logging.getLogger('numba').setLevel(logging.WARNING) |
|
logging.getLogger('markdown_it').setLevel(logging.WARNING) |
|
logging.getLogger('urllib3').setLevel(logging.WARNING) |
|
logging.getLogger('matplotlib').setLevel(logging.WARNING) |
|
|
|
config_path = "configs/config.json" |
|
|
|
model = Svc("logs/44k/G_130400.pth", "configs/config.json", cluster_model_path="logs/44k/kmeans.pt") |
|
|
|
|
|
|
|
def vc_fn(sid, input_audio, vc_transform, auto_f0,cluster_ratio, slice_db, noise_scale): |
|
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 > 90: |
|
return "请上传小于90s的音频，需要转换长音频请本地进行转换", 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) |
|
out_wav_path = "temp.wav" |
|
soundfile.write(out_wav_path, audio, 16000, format="wav") |
|
print( cluster_ratio, auto_f0, noise_scale) |
|
_audio = model.slice_inference(out_wav_path, sid, vc_transform, slice_db, cluster_ratio, auto_f0, noise_scale) |
|
return "Success", (44100, _audio) |
|
|
|
|
|
app = gr.Blocks() |
|
with app: |
|
with gr.Tabs(): |
|
with gr.TabItem("Basic"): |
|
gr.Markdown(value=""" |
|
# sovits-emu-voice-transform ｜ 可以变成凤笑梦的在线变声器 |
|
[](https://visitorbadge.io/status?path=https%3A%2F%2Fhuggingface.co%2Fspaces%2FMashiroSA%2Fsovits-emu-voice-transform) |
|
|
|
<br/> |
|
|
|
**说明 / Introduction** |
|
- 基于so-vits-svc 4.0的官方库示例修改而成。 |
|
- 该项目用于便携的基于云计算的变声成为Project Sekai的角色鳳えむ（凤笑梦）。 |
|
- 所使用的音声训练集基于对话而来，因而转换后的音声在对话表现中会比乐曲中的人声中要好。 |
|
- 该项目以无盈利模式进行。 |
|
- Modified from the official library example based on so-vits-svc 4.0. |
|
- The sound training set used is based on dialogue, thus the converted sound will perform better in dialogue than the vocals in the music. |
|
- This project is conducted in no-profit. |
|
```text |
|
For academic purpose only and not for illegal purposes. We have no relationship or interest with SEGA or related organizations. |
|
The model derivation output is only similar to Otori Emu and there is inevitable loss, which cannot be fully simulated. |
|
If you have any questions, please send an email or forum for inquiry. |
|
``` |
|
|
|
<br/> |
|
|
|
**如何使用** |
|
- 如果用于日常说话时的对话转换，请提前录制一段低于90s的人声干声，上传，勾选下面的自动f0预测，其它的可以不用动，直接转换，过一会儿就能听到转换的声音了。 |
|
- 如果是乐曲中的人声，你可以使用自己的清唱，或者使用UVR5软件进行干声提取，上传，不要勾选自动f0预测，按情况进行变调（模型实际测试高于标准音C4的类似度较高，输入的干声是男声请+12，女声可以先不变），然后转换。 |
|
- 转换后的进度条右侧有个省略的点，在那边可以下载。 |
|
- 本repo的管理者 @MashiroSA 看不到你输入和输出后的内容，只有Hugging Face官方也许可以看到，请放心。 |
|
- 关于下面选项中的聚类模型的使用：默认为0，值是0-1，越高越能贴近模型音色，但会导致咬字不清。 |
|
|
|
""") |
|
spks = list(model.spk2id.keys()) |
|
sid = gr.Dropdown(label="音色", choices=spks, value=spks[0]) |
|
vc_input3 = gr.Audio(label="上传音频（长度小于90秒）") |
|
vc_transform = gr.Number(label="变调（整数，可以正负，半音数量，升高八度就是12，当你觉得音色不准确时可以适当调高或降低，当自动f0预测勾选后该项失效）", value=0) |
|
cluster_ratio = gr.Number(label="聚类模型混合比例，0-1之间，默认为0不启用聚类，能提升音色相似度，但会导致咬字下降（如果使用建议0.5左右）", value=0) |
|
auto_f0 = gr.Checkbox(label="自动f0预测，配合聚类模型f0预测效果更好,会导致变调功能失效（仅限转换语音，歌声不要勾选此项会究极跑调）", value=False) |
|
slice_db = gr.Number(label="切片阈值", value=-40) |
|
noise_scale = gr.Number(label="noise_scale 建议不要动，会影响音质，玄学参数", value=0.4) |
|
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, [sid, vc_input3, vc_transform,auto_f0,cluster_ratio, slice_db, noise_scale], [vc_output1, vc_output2]) |
|
|
|
app.launch() |
|
|
|
|
|
|
|
|
