app.py

#!/usr/bin/env python3
#
# Copyright      2025  Xiaomi Corp.        (authors: Fangjun Kuang)
#
# See LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

# References:
# https://gradio.app/docs/#dropdown

import logging
import shutil
import tempfile
import time
import urllib.request
from datetime import datetime
import soundfile as sf

import gradio as gr
import numpy as np

from separate import get_file, load_audio, load_model, model_list

wav_files = [
    "yesterday-once-more-Carpenters.mp3",
    "das-beste-Silbermond.mp3",
    "hotel-in-california.mp3",
    "起风了.mp3",
]

for name in wav_files:
    filename = get_file(
        "csukuangfj/spleeter-torch",
        name,
        subfolder="test_wavs",
    )

    shutil.copyfile(filename, name)

examples = [[model_list[0], w] for w in wav_files]

logging.info(f"examples: {examples}")
print(f"examples: {examples}")


def build_html_output(s: str, style: str = "result_item_success"):
    return f"""
    <div class='result'>
        <div class='result_item {style}'>
          {s}
        </div>
    </div>
    """


def process_url(model_name: str, url: str):
    logging.info(f"Processing URL: {url}")
    with tempfile.NamedTemporaryFile() as f:
        try:
            urllib.request.urlretrieve(url, f.name)
            return process(model_name, in_filename=f.name)
        except Exception as e:
            logging.info(str(e))
            return "", "", build_html_output(str(e), "result_item_error")


def process_uploaded_file(model_name: str, in_filename: str):
    if in_filename is None or in_filename == "":
        return "", build_html_output(
            "Please first upload a file and then click "
            'the button "submit for separation"',
            "result_item_error",
        )

    logging.info(f"Processing uploaded file: {in_filename}")
    try:
        return process(model_name, in_filename=in_filename)
    except Exception as e:
        logging.info(str(e))
        return "", "", build_html_output(str(e), "result_item_error")


def process_microphone(model_name: str, in_filename: str):
    if in_filename is None or in_filename == "":
        return "", build_html_output(
            "Please first click 'Record from microphone', speak, "
            "click 'Stop recording', and then "
            "click the button 'submit for separation'",
            "result_item_error",
        )

    logging.info(f"Processing microphone: {in_filename}")
    try:
        return process(model_name, in_filename=in_filename)
    except Exception as e:
        logging.info(str(e))
        return "", "", build_html_output(str(e), "result_item_error")


def process(model_name, in_filename: str):
    logging.info(f"model_name: {model_name}")
    logging.info(f"in_filename: {in_filename}")

    samples, sample_rate = load_audio(in_filename)
    samples = np.ascontiguousarray(samples)

    duration = samples.shape[1] / sample_rate  # in seconds

    sp = load_model(model_name)

    now = datetime.now()
    date_time = now.strftime("%Y-%m-%d %H:%M:%S.%f")
    logging.info(f"Started at {date_time}")

    start = time.time()

    output = sp.process(sample_rate=sample_rate, samples=samples)

    date_time = now.strftime("%Y-%m-%d %H:%M:%S.%f")
    end = time.time()

    vocals = output.stems[0].data
    non_vocals = output.stems[1].data
    # vocals.shape (num_channels, num_samples)

    vocals = np.transpose(vocals)
    non_vocals = np.transpose(non_vocals)

    vocals_filename = in_filename + "-vocals.mp3"
    non_vocals_filename = in_filename + "-non-vocals.mp3"

    sf.write(vocals_filename, vocals, samplerate=output.sample_rate)
    sf.write(non_vocals_filename, non_vocals, samplerate=output.sample_rate)

    rtf = (end - start) / duration

    logging.info(f"Finished at {date_time} s. Elapsed: {end - start: .3f} s")

    info = f"""
    Input duration  : {duration: .3f} s <br/>
    Processing time: {end - start: .3f} s <br/>
    RTF: {end - start: .3f}/{duration: .3f} = {rtf:.3f} <br/>
    """
    logging.info(info)

    return vocals_filename, non_vocals_filename, build_html_output(info)


title = "# Source separation with Next-gen Kaldi"
description = """
This space shows how to do source separation with Next-gen Kaldi.

It is running on CPU within a docker container provided by Hugging Face.

See more information by visiting the following links:

- <https://github.com/k2-fsa/sherpa-onnx>

Everything is open-sourced.
"""

# css style is copied from
# https://huggingface.co/spaces/alphacep/asr/blob/main/app.py#L113
css = """
.result {display:flex;flex-direction:column}
.result_item {padding:15px;margin-bottom:8px;border-radius:15px;width:100%}
.result_item_success {background-color:mediumaquamarine;color:white;align-self:start}
.result_item_error {background-color:#ff7070;color:white;align-self:start}
"""


demo = gr.Blocks(css=css)


with demo:
    gr.Markdown(title)

    model_dropdown = gr.Dropdown(
        choices=model_list,
        label="Select a model",
        value=model_list[0],
    )

    with gr.Tabs():
        with gr.TabItem("Upload from disk"):
            uploaded_file = gr.Audio(
                sources=["upload"],  # Choose between "microphone", "upload"
                type="filepath",
                label="Upload from disk",
            )
            upload_button = gr.Button("Submit for separation")
            uploaded_html_info = gr.HTML(label="Info")

            uploaded_vocals = gr.Audio(label="vocals")
            uploaded_non_vocals = gr.Audio(label="non_vocals")

            gr.Examples(
                examples=examples,
                inputs=[model_dropdown, uploaded_file],
                outputs=[uploaded_vocals, uploaded_non_vocals, uploaded_html_info],
                fn=process_uploaded_file,
            )

        with gr.TabItem("Record from microphone"):
            microphone = gr.Audio(
                sources=["microphone"],  # Choose between "microphone", "upload"
                type="filepath",
                label="Record from microphone",
            )

            record_button = gr.Button("Submit for separation")
            recorded_html_info = gr.HTML(label="Info")

            recorded_vocals = gr.Audio(label="vocals")
            recorded_non_vocals = gr.Audio(label="non-vocals")

            gr.Examples(
                examples=examples,
                inputs=[model_dropdown, microphone],
                outputs=[recorded_vocals, recorded_non_vocals, recorded_html_info],
                fn=process_microphone,
            )

        with gr.TabItem("From URL"):
            url_textbox = gr.Textbox(
                max_lines=1,
                placeholder="URL to an audio file",
                label="URL",
                interactive=True,
            )

            url_button = gr.Button("Submit for separation")
            url_html_info = gr.HTML(label="Info")

            url_vocals = gr.Audio(label="vocals")
            url_non_vocals = gr.Audio(label="non-vocals")

            gr.Examples(
                examples=[
                    [
                        model_list[0],
                        "https://huggingface.co/csukuangfj/spleeter-torch/resolve/main/test_wavs/yesterday-once-more-Carpenters.mp3",
                    ],
                    [
                        model_list[0],
                        "https://huggingface.co/csukuangfj/spleeter-torch/resolve/main/test_wavs/das-beste-Silbermond.mp3",
                    ],
                    [
                        model_list[0],
                        "https://huggingface.co/csukuangfj/spleeter-torch/resolve/main/test_wavs/hotel-in-california.mp3",
                    ],
                ],
                inputs=[model_dropdown, url_textbox],
                outputs=[url_vocals, url_non_vocals, recorded_html_info],
                fn=process_url,
            )

        upload_button.click(
            process_uploaded_file,
            inputs=[model_dropdown, uploaded_file],
            outputs=[uploaded_vocals, uploaded_non_vocals, uploaded_html_info],
        )

        record_button.click(
            process_microphone,
            inputs=[model_dropdown, microphone],
            outputs=[recorded_vocals, recorded_non_vocals, recorded_html_info],
        )

        url_button.click(
            process_url,
            inputs=[model_dropdown, url_textbox],
            outputs=[url_vocals, url_non_vocals, url_html_info],
        )
    gr.Markdown(description)

if __name__ == "__main__":
    formatter = "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"

    logging.basicConfig(format=formatter, level=logging.INFO)

    demo.launch()