app.py

import gradio as gr
from infer_rvc_python import BaseLoader
import soundfile as sf
import random
from urllib.request import urlretrieve
import os
import zipfile

files_to_retrieve = [
    "https://replicate.delivery/pbxt/N97QM3XNFrooJhV6Fb0meBff0aAG1rEDfvuxcdLS6fTx1vmWC/test.zip",
    # "https://huggingface.co/lj1995/VoiceConversionWebUI/resolve/main/hubert_base.pt?download=true",
    # "https://huggingface.co/lj1995/VoiceConversionWebUI/resolve/main/rmvpe.pt?download=true"
]

for file in files_to_retrieve:
    print(f"Downloading {file}")
    urlretrieve(file, file.split("/")[-1])

# unzip test.zip
with zipfile.ZipFile("test.zip", "r") as zip_ref:
    zip_ref.extractall(".")


converter = BaseLoader(
    only_cpu=True, hubert_path="./hubert_base.pt", rmvpe_path="./rmvpe.pt"
)

model = "test.pth"
index = "added_IVF839_Flat_nprobe_1_test_v2.index"


def voice_conversion(
    audio,
    pitch_change,
    filter_radius,
    envelope_ratio,
    index_influence,
    consonant_breath_protection,
):
    global output_file
    audio_out = run(
        [str(audio)],
        model,
        "rmvpe+",
        pitch_change,
        index,
        index_influence,
        filter_radius,
        envelope_ratio,
        consonant_breath_protection,
    )
    print(audio_out)
    # output_audio, sr = sf.read(output_file, dtype="int32")
    return audio_out


def convert_now(audio_files, random_tag):
    return converter(audio_files, random_tag, overwrite=False, parallel_workers=8)


def run(
    audio_files,
    file_m,
    pitch_alg,
    pitch_lvl,
    file_index,
    index_inf,
    r_m_f,
    e_r,
    c_b_p,
):
    random_tag = "USER_" + str(random.randint(10000000, 99999999))
    print("PITCH LVL: ", pitch_lvl)
    converter.apply_conf(
        tag=random_tag,
        file_model=file_m,
        pitch_algo=pitch_alg,
        pitch_lvl=pitch_lvl,
        file_index=file_index,
        index_influence=index_inf,
        respiration_median_filtering=r_m_f,
        envelope_ratio=e_r,
        consonant_breath_protection=c_b_p,
        resample_sr=44100 if audio_files[0].endswith(".mp3") else 0,
    )
    output = convert_now(audio_files, random_tag)
    audio, sr = sf.read(output[0], dtype="int32")
    return (sr, audio)


def ui():
    with gr.Blocks() as demo:
        audio_input = gr.Audio(sources=["microphone", "upload"], type="filepath")
        with gr.Row():
            pitch_slider = gr.Slider(
                minimum=-24,
                maximum=24,
                value=0,
                step=1,
                label="Pitch",
                interactive=True,
            )
            index_influence_slider = gr.Slider(
                minimum=0,
                maximum=1,
                value=0.75,
                step=0.01,
                label="Index Influence",
                interactive=True,
            )
            respiration_median_filtering = gr.Slider(
                minimum=0,
                maximum=10,
                value=3,
                step=1,
                label="Resp. Median Filtering",
                interactive=True,
            )
            envelope_ratio = gr.Slider(
                minimum=0,
                maximum=1,
                value=0.25,
                step=0.01,
                label="Envelope Ratio",
                interactive=True,
            )
            consonant_breath_protection = gr.Slider(
                minimum=0,
                maximum=1,
                value=0.5,
                step=0.01,
                label="Consonant Breath Protection",
                interactive=True,
            )
        button = gr.Button("Convert")
        audio_output = gr.Audio(interactive=False, type="numpy")
        button.click(
            voice_conversion,
            inputs=[
                audio_input,
                pitch_slider,
                respiration_median_filtering,
                envelope_ratio,
                index_influence_slider,
                consonant_breath_protection,
            ],
            outputs=[audio_output],
        )

    return demo


ui().launch(auth=("output", "becreative"))