modules/tabs/inference.py

import glob
import os
import traceback

import gradio as gr

from modules import models, ui
from modules.ui import Tab


def inference_options_ui(show_out_dir=True):
    with gr.Row(equal_height=False):
        with gr.Column():
            source_audio = gr.Textbox(label="Source Audio")
            out_dir = gr.Textbox(
                label="Out folder",
                visible=show_out_dir,
                placeholder=models.AUDIO_OUT_DIR,
            )
        with gr.Column():
            transpose = gr.Slider(
                minimum=-20, maximum=20, value=0, step=1, label="Transpose"
            )
            pitch_extraction_algo = gr.Radio(
                choices=["dio", "harvest", "mangio-crepe", "crepe"],
                value="crepe",
                label="Pitch Extraction Algorithm",
            )
            embedding_model = gr.Radio(
                choices=["auto", *models.EMBEDDINGS_LIST.keys()],
                value="auto",
                label="Embedder Model",
            )
            embedding_output_layer = gr.Radio(
                choices=["auto", "9", "12"],
                value="auto",
                label="Embedder Output Layer",
            )
        with gr.Column():
            auto_load_index = gr.Checkbox(value=False, label="Auto Load Index")
            faiss_index_file = gr.Textbox(value="", label="Faiss Index File Path")
            retrieval_feature_ratio = gr.Slider(
                minimum=0,
                maximum=1,
                value=1,
                step=0.01,
                label="Retrieval Feature Ratio",
            )
        with gr.Column():
            fo_curve_file = gr.File(label="F0 Curve File")

    return (
        source_audio,
        out_dir,
        transpose,
        embedding_model,
        embedding_output_layer,
        pitch_extraction_algo,
        auto_load_index,
        faiss_index_file,
        retrieval_feature_ratio,
        fo_curve_file,
    )


class Inference(Tab):
    def title(self):
        return "Inference"

    def sort(self):
        return 1

    def ui(self, outlet):
        def infer(
            sid,
            input_audio,
            out_dir,
            embedder_model,
            embedding_output_layer,
            f0_up_key,
            f0_file,
            f0_method,
            auto_load_index,
            faiss_index_file,
            index_rate,
        ):
            model = models.vc_model
            try:
                yield "Infering...", None
                if out_dir == "":
                    out_dir = models.AUDIO_OUT_DIR

                if "*" in input_audio:
                    assert (
                        out_dir is not None
                    ), "Out folder is required for batch processing"
                    files = glob.glob(input_audio, recursive=True)
                elif os.path.isdir(input_audio):
                    assert (
                        out_dir is not None
                    ), "Out folder is required for batch processing"
                    files = glob.glob(
                        os.path.join(input_audio, "**", "*.wav"), recursive=True
                    )
                else:
                    files = [input_audio]
                for file in files:
                    audio = model.single(
                        sid,
                        file,
                        embedder_model,
                        embedding_output_layer,
                        f0_up_key,
                        f0_file,
                        f0_method,
                        auto_load_index,
                        faiss_index_file,
                        index_rate,
                        output_dir=out_dir,
                    )
                yield "Success", (model.tgt_sr, audio) if len(files) == 1 else None
            except:
                yield "Error: " + traceback.format_exc(), None

        with gr.Group():
            with gr.Box():
                with gr.Column():
                    _, speaker_id = ui.create_model_list_ui()

                    (
                        source_audio,
                        out_dir,
                        transpose,
                        embedder_model,
                        embedding_output_layer,
                        pitch_extraction_algo,
                        auto_load_index,
                        faiss_index_file,
                        retrieval_feature_ratio,
                        f0_curve_file,
                    ) = inference_options_ui()

                    with gr.Row(equal_height=False):
                        with gr.Column():
                            status = gr.Textbox(value="", label="Status")
                            output = gr.Audio(label="Output", interactive=False)

                    with gr.Row():
                        infer_button = gr.Button("Infer", variant="primary")

        infer_button.click(
            infer,
            inputs=[
                speaker_id,
                source_audio,
                out_dir,
                embedder_model,
                embedding_output_layer,
                transpose,
                f0_curve_file,
                pitch_extraction_algo,
                auto_load_index,
                faiss_index_file,
                retrieval_feature_ratio,
            ],
            outputs=[status, output],
            queue=True,
        )