app.py

import os
import io
import gradio as gr
import librosa
import numpy as np
import logging
import soundfile
import asyncio
import argparse
import gradio.processing_utils as gr_processing_utils
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)

limitation = os.getenv("SYSTEM") == "spaces"  # limit audio length in huggingface spaces

INFERENCE_OUTPUT_DIRNAME = '/output/'

audio_postprocess_ori = gr.Audio.postprocess

def audio_postprocess(self, y):
    data = audio_postprocess_ori(self, y)
    if data is None:
        return None
    return gr_processing_utils.encode_url_or_file_to_base64(data["name"])

gr.Audio.postprocess = audio_postprocess
def vc_fn(input_audio, vc_transform, voice):
    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 > 20 and limitation:
        return "Please upload an audio file that is less than 20 seconds. If you need to generate a longer audio file, please use 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)
    raw_path = io.BytesIO()
    soundfile.write(raw_path, audio, 16000, format="wav")
    raw_path.seek(0)
    out_audio, out_sr = model.infer(sid, vc_transform, raw_path,
                                   auto_predict_f0=True,
                                   )
    return "Success", (44100, out_audio.cpu().numpy())


def run_inference(input_audio, speaker):
    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 > 20 and limitation:
        return "Please upload an audio file that is less than 20 seconds. If you need to generate a longer audio file, please use 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)

    #TODO edit from GUI
    cluster_ratio = 1
    noise_scale = 2
    is_pitch_prediction_enabled = True
    f0_method = "dio"
    transpose = 0

    model_path = f"/models/{speaker}/{speaker}.pth"
    config_path = f"/models/{speaker}/config.json"
    cluster_path = ""

    raw_path = io.BytesIO()
    soundfile.write(raw_path, audio, 16000, format="wav")
    raw_path.seek(0)

    inference_cmd = f"svc infer {raw_path.absolute()} -m {model_path} -c {config_path} {f'-k {cluster_path} -r {cluster_ratio}' if cluster_path != '' and cluster_ratio > 0 else ''} -t {transpose} --f0-method {f0_method} -n {noise_scale} -o {INFERENCE_OUTPUT_DIRNAME}/{raw_path.name} {'' if is_pitch_prediction_enabled else '--no-auto-predict-f0'}"

    # out_audio, out_sr = model.infer(sid, vc_transform, raw_path,
    #                              auto_predict_f0=True,
    #                               )

    result = subprocess.run(
        inference_cmd.split(),
        stdout=subprocess.PIPE,
        stderr=subprocess.STDOUT,
        text=True
      )
    print(result)
    return "Success", "TODO" # (44100, out_audio.cpu().numpy())

if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--device', type=str, default='cpu')
    parser.add_argument('--api', action="store_true", default=False)
    parser.add_argument("--share", action="store_true", default=False, help="share gradio app")
    args = parser.parse_args()

    speakers = ["chapaev", "petka", "anka"]

    models = []
    voices = []

    # !svc infer {NAME}.wav -c config.json -m G_riri_220.pth
    #  display(Audio(f"{NAME}.out.wav", autoplay=True))
    with gr.Blocks() as app:
        gr.Markdown(
            "# <center> Sovits Chapay\n"
        )

        with gr.Row():
            with gr.Column():
                vc_input = gr.Audio(label="Input audio"+' (less than 20 seconds)' if limitation else '')
                speaker = gr.Dropdown(choices=speakers, visible=True)

                vc_submit = gr.Button("Generate", variant="primary")
            with gr.Column():
                vc_output1 = gr.Textbox(label="Output Message")
                # vc_output2 = gr.Audio(label="Output Audio")
            vc_submit.click(run_inference, [vc_input, speaker], [vc_output1, vc_output2])
        app.queue(concurrency_count=1, api_open=args.api).launch(share=args.share)