app.py

import av
import numpy as np
import torch
import streamlit as st
from PIL import Image
from torchvision.transforms import Resize, ToTensor, Compose
from transformers import AutoImageProcessor, AutoTokenizer, VisionEncoderDecoderModel
from diffusers import StableAudioPipeline
import soundfile as sf
import stripe
import time  # For simulating processing time
import firebase_admin
from firebase_admin import credentials, auth
from huggingface_hub import login
import os
import json

# Firebase JSON içeriğini Hugging Face Secrets'ten al
firebase_json_string = os.getenv("firebase")
firebase_credentials = json.loads(firebase_json_string)

hf_token = os.getenv("token1")  # Secret Manager'da tanımladığınız isim
login(hf_token)

# Firebase Admin SDK'yı başlatma
if not firebase_admin._apps:
    cred = credentials.Certificate(firebase_credentials)
    firebase_admin.initialize_app(cred)
# -----------------------------
# 1. Device Configuration
# -----------------------------
device = "cuda" if torch.cuda.is_available() else "cpu"

# -----------------------------
# 2. Streamlit Page Configuration and Styling
# -----------------------------
st.set_page_config(page_title="SoundScene.ai", page_icon=":musical_note:", layout="wide")

# Dark theme and custom styles
st.markdown(
    """
    <style>
        /* General Body Styles */
        body {
            background-color: #0e1117;
            color: #ffffff;
        }
        /* Title Styles */
        .stTitle {
            color: #1e90ff;
            text-align: center;
        }
        /* Sidebar Styles */
        .stSidebar {
            background-color: #1c1e26;
        }
        /* Button Styles */
        .stButton>button {
            background-color: #1e90ff;
            color: #ffffff;
        }
        .stButton>button:hover {
            background-color: #0d6efd;
        }
        /* Header Styles */
        .stHeader {
            color: #1e90ff;
        }
        /* Footer Styles */
        footer {
            visibility: hidden;
        }
    </style>
    """,
    unsafe_allow_html=True,
)

# -----------------------------
# 4. Stripe Configuration
# -----------------------------
# Replace with your actual Stripe secret key
stripe.api_key = "YOUR_STRIPE_SECRET_KEY"

# -----------------------------
# 5. Session State Initialization
# -----------------------------
# Initialize session states
if "user" not in st.session_state:
    st.session_state.user = None
if "auth_mode" not in st.session_state:
    st.session_state.auth_mode = "Login"  # Default mode must match radio option exactly
if "conversion_count" not in st.session_state:
    st.session_state.conversion_count = 0
if "subscribed" not in st.session_state:
    st.session_state.subscribed = False


# Authentication logic
def login_user(email, password):
    try:
        user = auth.get_user_by_email(email)
        # Note: Firebase Admin SDK does not handle password validation directly
        st.session_state.user = {"uid": user.uid, "email": user.email}
        st.success("Successfully logged in!")
    except Exception as e:
        st.error(f"Login failed: {str(e)}")


def register_user(email, password):
    try:
        user = auth.create_user(email=email, password=password)
        st.success("Registration successful! You can now log in.")
        st.session_state.auth_mode = "Login"  # Switch to login after registration
    except Exception as e:
        st.error(f"Registration failed: {str(e)}")


# Authentication interface
if st.session_state.user is None:
    #st.sidebar.title("Welcome to SoundScene.ai")
    st.sidebar.subheader("Please authenticate to use the application.")

    # Toggle between Login and Register
    auth_mode = st.sidebar.radio(
        "Select an option",
        ("Login", "Register"),
        key="auth_mode"
    )

    email = st.sidebar.text_input("Email", key="sidebar_email")
    password = st.sidebar.text_input("Password", type="password", key="sidebar_password")

    if auth_mode == "Login":
        if st.sidebar.button("Log In"):
            login_user(email, password)
    elif auth_mode == "Register":
        if st.sidebar.button("Register"):
            register_user(email, password)

# Main app (accessible regardless of login status)
#st.title("Welcome to SoundScene.ai")
if st.session_state.user:
    st.sidebar.success(f"Logged in as: {st.session_state.user['email']}")
    if st.sidebar.button("Log Out"):
        #st.session_state.clear()
        #st.experimental_rerun()  # Refresh app to show login screen
        st.session_state.user = None


@st.cache_resource
def load_models():
    audio_pipe = StableAudioPipeline.from_pretrained("stabilityai/stable-audio-open-1.0",use_auth_token=hf_token, torch_dtype=torch.float16)
    audio_pipe = audio_pipe.to(device)

    image_processor = AutoImageProcessor.from_pretrained("MCG-NJU/videomae-base")
    tokenizer = AutoTokenizer.from_pretrained("gpt2")
    video_model = VisionEncoderDecoderModel.from_pretrained("Neleac/timesformer-gpt2-video-captioning").to(device)

    return image_processor, tokenizer, video_model, audio_pipe


image_processor, tokenizer, video_model, audio_pipe = load_models()


# -----------------------------
# 7. Preprocessing Functions
# -----------------------------

# Görüntüleri yeniden boyutlandıran bir işlev
def preprocess_frame(frame, target_size=(224, 224)):
    """
    Preprocess a single video frame for model input.
    Args:
        frame (np.ndarray): Video frame.
        target_size (tuple): Desired size (height, width).
    Returns:
        PIL.Image: Preprocessed frame image.
    """
    # Ensure the frame is in uint8 format
    if frame.dtype != np.uint8:
        frame = np.clip(frame, 0, 255).astype(np.uint8)

    # Check frame dimensions
    if frame.ndim != 3 or frame.shape[2] != 3:
        raise ValueError(f"Expected frame with 3 channels, got shape {frame.shape}")

    transform = Resize(target_size)
    frame_image = Image.fromarray(frame)
    frame_image = transform(frame_image)  # Resize the image
    return frame_image  # Returns PIL.Image


def preprocess_frames(frames, target_size=(224, 224), num_frames=8):
    """
    Preprocess multiple video frames for model input.
    Args:
        frames (list of np.ndarray): Video frames.
        target_size (tuple): Desired size (height, width).
        num_frames (int): Number of frames to process.
    Returns:
        list of PIL.Image: List of preprocessed frame images.
    """
    processed_frames = []
    for idx, frame in enumerate(frames[:num_frames]):
        try:
            img = preprocess_frame(frame, target_size)
            # Debugging: Display frame properties
            st.write(f"Frame {idx + 1}: size={img.size}, mode={img.mode}")
            processed_frames.append(img)
        except Exception as e:
            st.warning(f"Skipping frame {idx + 1} due to error: {e}")
    return processed_frames


# with st.sidebar:
#     st.title("Login")
#     email = st.text_input("Email", key="login_email")
#     password = st.text_input("Password", type="password", key="login_password")
#     login_btn = st.button("Login")
#     logout_btn = st.button("Logout")

#     if login_btn:
#         try:
#             user = auth.sign_in_with_email_and_password(email, password)
#             st.session_state.user = user
#             st.success("Logged in successfully!")
#         except:
#             st.error("Invalid login credentials.")
#     elif logout_btn:
#         st.session_state.user = None
#         st.success("Logged out successfully!")

#     if st.session_state.user:
#         st.sidebar.write(f"Welcome, {st.session_state.user['email']}!")

# -----------------------------
# 9. Navigation Menu
# -----------------------------
with st.sidebar:
    st.markdown("---")
    page = st.selectbox("Menu", ["Home", "About", "Pricing", "Contact"])


# -----------------------------
# 10. Membership Control and Payment Integration
# -----------------------------
def handle_subscription():
    st.warning("Subscription service is currently not available.")


# -----------------------------
# 11. Main Content Rendering Based on Navigation
# -----------------------------
if page == "Home":
    # -----------------------------
    # 12. Home Page Content
    # -----------------------------
    st.title("Welcome to SoundScene.ai")
    st.markdown(
        """
        <p style='text-align: center;'>
            Transform your videos into stunning soundscapes with advanced AI models. 
            Learn more about our <a href="#About" style="color: #1e90ff;">features</a> and <a href="#Pricing" style="color: #1e90ff;">pricing</a> options.
        </p>
        """,
        unsafe_allow_html=True
    )

    st.header("Create Audio")
    creation_method = st.radio(
        "Choose how you want to create audio:",
        ("Upload Video", "Input Comment"),
        horizontal=True
    )

    # Initialize variables
    caption = ""
    audio_path = ""
    audio_duration = 0

    # -----------------------------
    # 13. Audio Creation via Comment
    # -----------------------------
    if creation_method == "Input Comment":
        st.subheader("Create Audio via Comment")
        user_comment = st.text_input("Enter your comment here:")

        # Sound Duration Input
        default_duration = 10  # Default to 10 seconds
        max_duration = 60  # Maximum allowed duration
        user_duration = st.number_input(
            "Select sound duration (seconds):",
            min_value=1,
            max_value=max_duration,
            value=default_duration,
            step=1
        )

        if st.button("Generate Sound"):
            if not user_comment:
                st.error("Please enter a comment to generate sound.")
            else:
                try:
                    # Initialize progress bar
                    progress_bar = st.progress(0)
                    progress_text = st.empty()

                    with st.spinner("Generating sound from your comment..."):
                        # Step 1: Generating audio
                        progress_text.text("Generating audio...")
                        # Simulate processing time
                        time.sleep(1)
                        progress_bar.progress(20)

                        # Generate audio based on comment
                        negative_prompt = "Low quality."
                        generator = torch.Generator(device).manual_seed(0)
                        audio = audio_pipe(
                            prompt=user_comment,
                            negative_prompt=negative_prompt,
                            num_inference_steps=50,
                            audio_end_in_s=user_duration,
                            num_waveforms_per_prompt=1,
                            generator=generator,
                        ).audios

                        progress_bar.progress(60)
                        progress_text.text("Finalizing audio...")

                        # Save audio
                        output = audio[0].T.float().cpu().numpy()
                        audio_path = "generated_audio.wav"
                        sf.write(audio_path, output, audio_pipe.vae.sampling_rate)

                        progress_bar.progress(80)
                        progress_text.text("Almost done...")

                        # Update conversion count
                        st.session_state.conversion_count += 1

                        # Finalize progress bar
                        progress_bar.progress(100)
                        progress_text.text("Audio generation complete!")

                        # Display audio
                        st.audio(audio_path)

                        # Download button logic
                        if st.session_state.user:
                            if st.session_state.subscribed or st.session_state.conversion_count <= 10:
                                st.download_button("Download Sound", audio_path, file_name="soundscene_output.wav")
                            else:
                                st.warning(
                                    "You have reached the download limit. Please subscribe for unlimited downloads.")
                        else:
                            st.warning("Please log in to download the generated sound.")

                        # Clear progress text after a short delay
                        time.sleep(1)
                        progress_text.empty()
                        progress_bar.empty()

                except Exception as e:
                    st.error(f"An error occurred while generating audio: {e}")

    # -----------------------------
    # 14. Audio Creation via Video Upload
    # -----------------------------
    elif creation_method == "Upload Video":
        st.subheader("Create Audio via Video Upload")
        uploaded_video = st.file_uploader("Upload a video", type=["mp4", "avi", "mov"])

        if uploaded_video:
            try:
                # Open the video using PyAV
                container = av.open(uploaded_video)
                video_stream = container.streams.video[0]
                video_duration = float(container.duration * video_stream.time_base)
                st.write(f"Video Duration: {video_duration:.2f} seconds")

                # Sound Duration Selection
                sound_duration = st.slider(
                    "Select sound duration (seconds):",
                    min_value=1,
                    max_value=10,
                    value=int(video_duration),
                    step=1
                )

                if st.button("Process Video"):
                    try:
                        # Initialize progress bar
                        progress_bar = st.progress(0)
                        progress_text = st.empty()

                        with st.spinner("Processing your video..."):
                            # Step 1: Extracting frames
                            progress_text.text("Extracting frames from video...")
                            seg_len = video_stream.frames
                            clip_len = video_model.config.encoder.num_frames
                            if clip_len > seg_len:
                                st.warning(f"Video has only {seg_len} frames, but the model expects {clip_len} frames.")
                                clip_len = seg_len
                            # Simulate processing time
                            time.sleep(1)
                            progress_bar.progress(10)

                            # Select evenly spaced frame indices
                            indices = set(np.linspace(0, seg_len - 1, num=clip_len, endpoint=True).astype(np.int64))

                            frames = []
                            container.seek(0)
                            for i, frame in enumerate(container.decode(video=0)):
                                if i in indices:
                                    frames.append(frame.to_ndarray(format="rgb24"))

                            if not frames:
                                st.error("No frames were extracted from the video.")
                                st.stop()

                            progress_bar.progress(30)
                            progress_text.text("Preprocessing frames...")

                            # Preprocess frames
                            processed_frames = preprocess_frames(frames, target_size=(224, 224), num_frames=clip_len)
                            if not processed_frames:
                                st.error("No valid frames to process after preprocessing.")
                                st.stop()

                            # Simulate processing time
                            time.sleep(1)
                            progress_bar.progress(50)

                            # Step 2: Generating caption
                            progress_text.text("Preprocessing the video...")
                            # Ensure image_processor receives correct keyword argument
                            pixel_values = image_processor(images=processed_frames,
                                                           return_tensors="pt").pixel_values.to(device)
                            gen_kwargs = {"min_length": 10, "max_length": 20, "num_beams": 8}
                            tokens = video_model.generate(pixel_values, **gen_kwargs)
                            caption = tokenizer.batch_decode(tokens, skip_special_tokens=True)[0]
                            st.write(f"Generated Caption: {caption}")
                            # Simulate processing time
                            time.sleep(1)
                            progress_bar.progress(70)
                            if sound_duration > 10:
                                sound_duration = 10
                            # Step 3: Generating audio based on caption
                            progress_text.text("Generating sound based on the video...")
                            negative_prompt = "Low quality."
                            generator = torch.Generator(device).manual_seed(0)
                            audio = audio_pipe(
                                prompt=caption,
                                negative_prompt=negative_prompt,
                                num_inference_steps=50,
                                audio_end_in_s=sound_duration,
                                num_waveforms_per_prompt=1,
                                generator=generator,
                            ).audios

                            progress_bar.progress(90)
                            progress_text.text("Finalizing audio...")

                            # Save audio
                            output = audio[0].T.float().cpu().numpy()
                            audio_path = "generated_audio.wav"
                            sf.write(audio_path, output, audio_pipe.vae.sampling_rate)

                            # Update conversion count
                            st.session_state.conversion_count += 1

                            # Finalize progress bar
                            progress_bar.progress(100)
                            progress_text.text("Audio generation complete!")

                            # Display audio
                            st.audio(audio_path)

                            # Download button logic
                            if st.session_state.user:
                                if st.session_state.subscribed or st.session_state.conversion_count <= 10:
                                    st.download_button("Download Sound", audio_path, file_name="soundscene_output.wav")
                                else:
                                    st.warning(
                                        "You have reached the download limit. Please subscribe for unlimited downloads.")
                            else:
                                st.warning("Please log in to download the generated sound.")

                            # Clear progress text and bar after a short delay
                            time.sleep(1)
                            progress_text.empty()
                            progress_bar.empty()

                    except Exception as e:
                        st.error(f"An error occurred while processing the video: {e}")
            except Exception as e:
                st.error(f"Failed to open the video file: {e}")

    # -----------------------------
    # 15. Conversion Count and Subscription Prompt
    # -----------------------------
    if st.session_state.conversion_count >= 10 and not st.session_state.subscribed:
        st.warning("You have used your 10 free conversions. Please subscribe to process more content.")
        if st.button("Subscribe for $5/month (Currently not available)"):
            handle_subscription()

elif page == "About":
    # -----------------------------
    # 16. About Page Content
    # -----------------------------
    st.title("About SoundScene.ai")
    st.markdown(
        """
        <p style='text-align: center;'>
            SoundScene.ai leverages cutting-edge AI technologies to transform your visual content into immersive soundscapes. Whether you're a content creator, filmmaker, or enthusiast, our platform empowers you to add rich audio dimensions to your videos effortlessly.
        </p>
        <br>
        <h3>Features:</h3>
        <ul>
            <li>Generate audio from video content with precise synchronization.</li>
            <li>Create custom soundscapes based on textual descriptions.</li>
            <li>Download and share your unique audio creations.</li>
            <li>Secure subscription plans to cater to your needs.</li>
        </ul>
        <br>
        <h3>Base Models From Huggingface:</h3>
        <p>
                <li>Image Processing: MCG-NJU/videomae-base</li>
                <li>Tokenizer: gpt2</li>
                <li>Video Model: Neleac/timesformer-gpt2-video-captioning</li>
                <li>Audio Model = stabilityai/stable-audio-open-1.0</li>


        </p>

        <a href="https://huggingface.co/stabilityai/stable-audio-open-1.0" target="_blank" style="color: #1e90ff;"> stable-audio-open-1.0 </a>

        <a href="https://huggingface.co/Neleac/SpaceTimeGPT" target="_blank" style="color: #1e90ff;"> SpaceTimeGPT </a>
        """,
        unsafe_allow_html=True
    )

elif page == "Pricing":
    # -----------------------------
    # 17. Pricing Page Content
    # -----------------------------
    st.title("Pricing")
    st.markdown(
        """
        <p style='text-align: center;'>
            Choose a plan that fits your needs and start transforming your content today!
        </p>
        """,
        unsafe_allow_html=True
    )

    # Pricing Cards
    col1, col2, col3 = st.columns(3)

    with col1:
        st.markdown(
            """
            <div style="background-color:#1c1e26;padding:20px;border-radius:10px;">
                <h3 style="text-align:center;">Free</h3>
                <p style="text-align:center; font-size:24px;">$0/month</p>
                <ul style="color:#ffffff;">
                    <li>Up to 10 conversions per month</li>
                    <li>Basic Support</li>
                </ul>
                <div style="text-align:center;">
                    <button style="background-color:#1e90ff; color:#ffffff; padding:10px 20px; border:none; border-radius:5px;">Select</button>
                </div>
            </div>
            """,
            unsafe_allow_html=True
        )

    with col2:
        st.markdown(
            """
            <div style="background-color:#1c1e26;padding:20px;border-radius:10px;">
                <h3 style="text-align:center;">Pro</h3>
<p style="text-align:center; font-size:24px;">$5/month (Currently not available)</p>
                <ul style="color:#ffffff;">
                    <li>Unlimited conversions</li>
                    <li>Priority Support</li>
                    <li>Access to Premium Features</li>
                </ul>
                <div style="text-align:center;">
                    <button style="background-color:#1e90ff; color:#ffffff; padding:10px 20px; border:none; border-radius:5px;" onclick="window.location.href='#'">Select</button>
                </div>
            </div>
            """,
            unsafe_allow_html=True
        )

    with col3:
        st.markdown(
            """
            <div style="background-color:#1c1e26;padding:20px;border-radius:10px;">
                <h3 style="text-align:center;">Enterprise</h3>
                <p style="text-align:center; font-size:24px;">Contact Us</p>
                <ul style="color:#ffffff;">
                    <li>Custom Solutions</li>
                    <li>Dedicated Support</li>
                    <li>Scalable Infrastructure</li>
                </ul>
                <div style="text-align:center;">
                    <button style="background-color:#1e90ff; color:#ffffff; padding:10px 20px; border:none; border-radius:5px;" onclick="window.location.href='#'">Contact</button>
                </div>
            </div>
            """,
            unsafe_allow_html=True
        )

elif page == "Contact":
    # -----------------------------
    # 18. Contact Page Content
    # -----------------------------
    st.title("Contact Us")
    st.markdown(
        """
        <p style='text-align: center;'>
            We'd love to hear from you! Whether you have a question about features, trials, pricing, or anything else, our team is ready to answer all your questions.
        </p>
        """,
        unsafe_allow_html=True
    )

    # Contact Form
    with st.form("contact_form"):
        name = st.text_input("Your Name")
        email = st.text_input("Your Email")
        message = st.text_area("Your Message")
        submitted = st.form_submit_button("Send")
        if submitted:
            if not name or not email or not message:
                st.error("Please fill out all fields.")
            else:
                # Here you can integrate with an email service or database to store the messages
                st.success("Thank you for reaching out! We'll get back to you shortly.")

    # Contact Details
    st.markdown("---")
    st.markdown(
        """
        <h3>Social Media Accounts</h3>
        <a href="https://www.linkedin.com/in/yunusakkaya/" target="_blank" style="color: #1e90ff;"> Linkedin </a>

        <a href="https://medium.com/@yunus-akkaya" target="_blank" style="color: #1e90ff;"> Medium </a>

        """,
        unsafe_allow_html=True
    )

# -----------------------------
# 19. Conversion Count and Subscription Prompt (Moved to Home Page)
# -----------------------------
# Note: This section is already handled within the "Home" page based on conversion count.

# -----------------------------
# 20. Security and Best Practices Notes
# -----------------------------
# Note:
# - Replace placeholder API keys with secure methods (e.g., environment variables or Streamlit secrets).
# - Ensure proper configuration of Firebase and Stripe if enabling authentication and payment features.
# - Validate and sanitize all user inputs to enhance security.
# - Customize the success_url and cancel_url in the Stripe checkout session to match your deployment URLs.

# -----------------------------
# 21. Footer (Hidden)
# -----------------------------
# Optional: Add a custom footer if desired
st.markdown(
    """
    <style>
        .footer {
            position: fixed;
            left: 0;
            bottom: 0;
            width: 100%;
            background-color: #1c1e26;
            color: white;
            text-align: center;
            padding: 10px 0;
        }
    </style>
    <div class="footer">
        <p>© 2024 SoundScene.ai. All rights reserved.</p>
    </div>
    """,
    unsafe_allow_html=True
)