src/streamlit_app.py

import streamlit as st
import numpy as np
import librosa
import matplotlib.pyplot as plt
from sklearn.decomposition import FastICA
import soundfile as sf
import io
import cv2
from skimage import io as skio

def custom_styles():
    """Apply custom CSS styles for the Streamlit app."""
    st.markdown("""
    <style>
    .big-font {
        font-size:18px !important;
    }
    .signature {
        font-size:16px;
        font-weight:bold;
    }
    </style>
    """, unsafe_allow_html=True)

def reconstruct_image_ica(image, n_components):
    """Perform ICA and reconstruct the image based on a given number of components."""
    reshaped_image = image.reshape(-1, 3)
    ica = FastICA(n_components=n_components, random_state=0, whiten='unit-variance', max_iter=500)
    ica.fit(reshaped_image)
    transformed_ica = ica.transform(reshaped_image)
    restored_image = ica.inverse_transform(transformed_ica)
    restored_image = restored_image.reshape(image.shape)
    restored_image = np.clip(restored_image, 0, 255)
    return restored_image

def load_color_image_from_upload(file):
    """Load a color image from an uploaded file."""
    bytes_data = file.getvalue()
    image = skio.imread(io.BytesIO(bytes_data))
    return image

def load_audio_from_upload(file, sr=22050, duration=30):
    """Load audio from an uploaded file."""
    bytes_data = file.getvalue()
    audio, _ = librosa.load(io.BytesIO(bytes_data), sr=sr, duration=duration)
    return audio

def mix_audios(audio1, audio2, weights):
    """Mix two audio signals with given weights."""
    return audio1 * weights[0] + audio2 * weights[1]

def separate_audio(mixed_signals, n_components=2):
    """Separate mixed audio signals using ICA."""
    ica = FastICA(n_components=n_components, random_state=0)
    separated_signals = ica.fit_transform(mixed_signals.T).T
    return separated_signals

def plot_signals(audios, titles, sr=22050):
    """Plot and display audio signals."""
    fig, axs = plt.subplots(len(audios), 1, figsize=(10, 2 * len(audios)))
    if len(audios) == 1:
        axs = [axs]
    times = np.linspace(0, len(audios[0]) / sr, num=len(audios[0]))
    for i, (audio, title) in enumerate(zip(audios, titles)):
        axs[i].plot(times, audio)
        axs[i].set_title(title)
    plt.tight_layout()
    st.pyplot(fig)
    plt.close(fig)

def display_audio_button(audio, title, sr=22050):
    """Display an audio player inside an expander."""
    with st.expander(f"Click here to listen to {title}"):
        audio_buffer = io.BytesIO()
        sf.write(audio_buffer, audio, sr, format='wav')
        audio_buffer.seek(0)
        st.audio(audio_buffer, format='audio/wav')

def load_image_from_upload(file):
    """Load a grayscale image from an uploaded file."""
    bytes_data = file.getvalue()
    image = cv2.imdecode(np.frombuffer(bytes_data, np.uint8), cv2.IMREAD_GRAYSCALE)
    return image

def separate_images(image1, image2, n_components=2):
    """Separate mixed images using ICA."""
    S1 = image1
    S2 = cv2.resize(image2, (S1.shape[1], S1.shape[0]))
    w = np.array([[0.6, 0.4], [0.4, 0.6]])
    X1 = w[0, 0] * S1 + w[0, 1] * S2
    X2 = w[1, 0] * S1 + w[1, 1] * S2
    stacked_images = np.vstack((X1.flatten(), X2.flatten())).T
    ica = FastICA(n_components=n_components, max_iter=1000, tol=0.1)
    transformed_sources = ica.fit_transform(stacked_images).T
    separated_img1 = (
        (transformed_sources[0] - transformed_sources[0].min())
        * (255 / (transformed_sources[0].max() - transformed_sources[0].min()))
    ).astype(np.uint8).reshape(S1.shape)
    separated_img2 = (
        (transformed_sources[1] - transformed_sources[1].min())
        * (255 / (transformed_sources[1].max() - transformed_sources[1].min()))
    ).astype(np.uint8).reshape(S2.shape)
    return separated_img1, separated_img2, X1.astype(np.uint8), X2.astype(np.uint8)

def display_image(image, title):
    """Display an image with a title."""
    st.image(image, caption=title, use_container_width=True)

def main():
    st.set_page_config(
        page_title="Cocktail Party ICA",
        page_icon="🎵",
        layout="wide",
    )
    custom_styles()

    st.title("🍹 Welcome to the Cocktail Party (18+)!")
    st.markdown("#### Choose the process: separate audio/images or reconstruct images using ICA.")

    process_choice = st.radio(
        "",
        ("Audio Separation", "Image Separation", "Image Reconstruction"),
        horizontal=True,
    )

    # ── Audio Separation ──────────────────────────────────────────────────────
    if process_choice == "Audio Separation":
        st.header("🎸🎺🥁🎵 Audio Separation")
        audio1_file = st.file_uploader(
            "Upload first audio file (.mp3, .wav)", type=["mp3", "wav"], key="audio1"
        )
        audio2_file = st.file_uploader(
            "Upload second audio file (.mp3, .wav)", type=["mp3", "wav"], key="audio2"
        )

        if audio1_file and audio2_file:
            n_components_audio = st.number_input(
                "Number of ICA components for audio",
                min_value=1, max_value=10, value=2, step=1, key="n_components_audio",
            )

            if st.button("Process Audios"):
                with st.spinner("Running ICA on audio signals…"):
                    audio1 = load_audio_from_upload(audio1_file)
                    audio2 = load_audio_from_upload(audio2_file)

                    # Trim to same length
                    min_len = min(len(audio1), len(audio2))
                    audio1, audio2 = audio1[:min_len], audio2[:min_len]

                    mixed_audio1 = mix_audios(audio1, audio2, [0.6, 0.4])
                    mixed_audio2 = mix_audios(audio1, audio2, [0.5, 0.5])
                    mixed_signals = np.vstack([mixed_audio1, mixed_audio2])
                    separated_audios = separate_audio(mixed_signals, n_components=int(n_components_audio))

                    audios = [audio1, audio2, mixed_audio1, mixed_audio2,
                              separated_audios[0], separated_audios[1]]
                    titles = [
                        "Original Audio 1", "Original Audio 2",
                        "Mixed Audio 1", "Mixed Audio 2",
                        "Separated Audio 1", "Separated Audio 2",
                    ]
                    for audio, title in zip(audios, titles):
                        plot_signals([audio], [title])
                        display_audio_button(audio, title)

    # ── Image Separation ──────────────────────────────────────────────────────
    elif process_choice == "Image Separation":
        st.header("🌲🖼️🏯🏰 Image Separation")
        image1_file = st.file_uploader(
            "Upload first image file", type=["jpg", "png"], key="image1"
        )
        image2_file = st.file_uploader(
            "Upload second image file", type=["jpg", "png"], key="image2"
        )

        if image1_file and image2_file:
            n_components_image = st.number_input(
                "Number of ICA components for images",
                min_value=1, max_value=10, value=2, step=1, key="n_components_image",
            )
            if st.button("Process Images"):
                with st.spinner("Running ICA on images…"):
                    image1 = load_image_from_upload(image1_file)
                    image2 = load_image_from_upload(image2_file)
                    separated_img1, separated_img2, mixed_img1, mixed_img2 = separate_images(
                        image1, image2, n_components=int(n_components_image)
                    )

                col1, col2 = st.columns(2)
                with col1:
                    display_image(image1, "Original Image 1")
                    display_image(mixed_img1, "Mixed Image 1")
                    display_image(separated_img1, "Separated Image 1")
                with col2:
                    display_image(image2, "Original Image 2")
                    display_image(mixed_img2, "Mixed Image 2")
                    display_image(separated_img2, "Separated Image 2")

    # ── Image Reconstruction ──────────────────────────────────────────────────
    elif process_choice == "Image Reconstruction":
        st.header("🎨 Image Reconstruction")
        image_file = st.file_uploader(
            "Upload an image file", type=["jpg", "png", "jpeg"], key="reconstruct_image"
        )
        if image_file:
            n_components_image = st.number_input(
                "Number of ICA components for reconstruction",
                min_value=1, max_value=3, value=1, step=1, key="n_components_reconstruct",
            )
            if st.button("Reconstruct Image"):
                with st.spinner("Reconstructing image with ICA…"):
                    color_image = load_color_image_from_upload(image_file)
                    # Drop alpha channel if present (e.g. PNG with transparency)
                    if color_image.shape[-1] == 4:
                        color_image = color_image[..., :3]
                    reconstructed_image = reconstruct_image_ica(
                        color_image, int(n_components_image)
                    )

                col1, col2 = st.columns(2)
                with col1:
                    st.image(color_image, caption="Original Image", use_container_width=True)
                with col2:
                    st.image(
                        reconstructed_image.astype(np.uint8),
                        caption=f"Reconstructed with {n_components_image} component(s)",
                        use_container_width=True,
                    )

    st.markdown("---")
    st.markdown('<p class="signature">Created by Dr. Jishan Ahmed</p>', unsafe_allow_html=True)

if __name__ == "__main__":
    main()