new

app.py

@@ -5,12 +5,7 @@ import numpy as np
 import librosa
 import cv2
 import re
-from transformers import (
-    Wav2Vec2Processor,
-    Wav2Vec2Model,
-    AutoTokenizer,
-    AutoModel
-)
+from transformers import Wav2Vec2Processor, Wav2Vec2Model, AutoTokenizer, AutoModel
 from torchvision import models
 import tempfile
 import os
@@ -18,391 +13,187 @@ from huggingface_hub import hf_hub_download
 import whisper
 import subprocess
 
-# =========================================================
-# CONFIGURATION
-# =========================================================
-
+# Configuration
 DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-
 SAMPLE_RATE = 16000
-
 TEXT_MAX_LEN = 64
-
 LABELS = ["angry", "happy", "neutral", "sad"]
 
-gpu_status = (
-    "GPU ONLINE"
-    if torch.cuda.is_available()
-    else "CPU MODE"
-)
-
-gpu_color = "#00ff88" if torch.cuda.is_available() else "#ff6b35"
-
-# =========================================================
-# LOAD PROCESSORS
-# =========================================================
-
-processor = Wav2Vec2Processor.from_pretrained(
-    "facebook/wav2vec2-base-960h"
-)
-
-tokenizer = AutoTokenizer.from_pretrained(
-    "bert-base-uncased"
-)
-
-# =========================================================
-# MODEL ARCHITECTURE
-# =========================================================
+# Load processors
+processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base-960h")
+tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
 
+# Model Architecture (same as training)
 class ResNetVideoEncoder(nn.Module):
-
     def __init__(self, out_dim=768):
-
         super().__init__()
-
         base = models.resnet18(pretrained=False)
-
-        self.backbone = nn.Sequential(
-            *list(base.children())[:-1]
-        )
-
+        self.backbone = nn.Sequential(*list(base.children())[:-1])
         self.proj = nn.Linear(512, out_dim)
 
     def forward(self, x):
-
         B, C, T, H, W = x.shape
-
         feats = []
-
         for t in range(T):
-
             ft = self.backbone(x[:, :, t])
-
-            feats.append(
-                ft.squeeze(-1).squeeze(-1)
-            )
-
+            feats.append(ft.squeeze(-1).squeeze(-1))
         feats = torch.stack(feats, dim=1).mean(1)
-
         return self.proj(feats)
 
-# =========================================================
-
 def mean_pool(x, mask):
-
     mask = mask[:, :x.size(1)]
-
     mask = mask.unsqueeze(-1).float()
-
-    return (
-        (x * mask).sum(1)
-        / mask.sum(1).clamp(min=1e-6)
-    )
-
-# =========================================================
+    return (x * mask).sum(1) / mask.sum(1).clamp(min=1e-6)
 
 class HBF(nn.Module):
-
     def __init__(self, d=768, n_layers=6):
-
         super().__init__()
-
-        self.proj_a = nn.ModuleList(
-            [nn.Linear(d, d) for _ in range(n_layers)]
-        )
-
-        self.proj_t = nn.ModuleList(
-            [nn.Linear(d, d) for _ in range(n_layers)]
-        )
-
-        self.proj_v = nn.ModuleList(
-            [nn.Linear(d, d) for _ in range(n_layers)]
-        )
-
-        self.fwd1 = nn.ModuleList(
-            [nn.Linear(3*d, d) for _ in range(n_layers)]
-        )
-
-        self.fwd2 = nn.ModuleList(
-            [nn.Linear(d, d) for _ in range(n_layers)]
-        )
-
+        self.proj_a = nn.ModuleList([nn.Linear(d, d) for _ in range(n_layers)])
+        self.proj_t = nn.ModuleList([nn.Linear(d, d) for _ in range(n_layers)])
+        self.proj_v = nn.ModuleList([nn.Linear(d, d) for _ in range(n_layers)])
+        self.fwd1 = nn.ModuleList([nn.Linear(3*d, d) for _ in range(n_layers)])
+        self.fwd2 = nn.ModuleList([nn.Linear(d, d) for _ in range(n_layers)])
         self.drop = nn.Dropout(0.1)
-
-        self.act1 = nn.GELU()
-
-        self.act2 = nn.Tanh()
-
+        self.act1, self.act2 = nn.GELU(), nn.Tanh()
         self.n = n_layers
 
     def forward(self, a, t, v):
-
         v_prev = None
-
         for i in range(self.n):
-
-            va = self.act2(
-                self.drop(self.proj_a[i](a))
-            )
-
-            vt = self.act2(
-                self.drop(self.proj_t[i](t))
-            )
-
-            vv = self.act2(
-                self.drop(self.proj_v[i](v))
-            )
-
-            cat = torch.cat(
-                [va, vt, vv]
-                if v_prev is None
-                else [va, vt, v_prev],
-                -1
-            )
-
+            va = self.act2(self.drop(self.proj_a[i](a)))
+            vt = self.act2(self.drop(self.proj_t[i](t)))
+            vv = self.act2(self.drop(self.proj_v[i](v)))
+            cat = torch.cat([va, vt, vv] if v_prev is None else [va, vt, v_prev], -1)
             x = self.act1(self.fwd1[i](cat))
-
             v_prev = self.fwd2[i](x)
-
         return v_prev
 
-# =========================================================
-
 class AVVideoModel(nn.Module):
-
     def __init__(self, num_classes, n_layers=6):
-
         super().__init__()
-
-        self.a_enc = Wav2Vec2Model.from_pretrained(
-            "facebook/wav2vec2-base-960h"
-        )
-
-        self.t_enc = AutoModel.from_pretrained(
-            "bert-base-uncased"
-        )
-
+        self.a_enc = Wav2Vec2Model.from_pretrained("facebook/wav2vec2-base-960h")
+        self.t_enc = AutoModel.from_pretrained("bert-base-uncased")
         self.v_enc = ResNetVideoEncoder()
-
         self.hbf = HBF(n_layers=n_layers)
-
         self.fc = nn.Linear(768, num_classes)
+        self.fc_audio = nn.Linear(768, num_classes)
+        self.fc_text = nn.Linear(768, num_classes)
+        self.fc_video = nn.Linear(768, num_classes)
 
-    def forward(
-        self,
-        audio,
-        audio_mask,
-        text_ids,
-        text_mask,
-        video
-    ):
-
-        a_out = self.a_enc(
-            audio,
-            attention_mask=audio_mask,
-            return_dict=True
-        )
-
-        t_out = self.t_enc(
-            input_ids=text_ids,
-            attention_mask=text_mask,
-            return_dict=True
-        )
-
-        a_pool = mean_pool(
-            a_out.last_hidden_state,
-            audio_mask
-        )
-
-        t_pool = mean_pool(
-            t_out.last_hidden_state,
-            text_mask
-        )
-
+    def forward(self, audio, audio_mask, text_ids, text_mask, video):
+        a_out = self.a_enc(audio, attention_mask=audio_mask, return_dict=True)
+        t_out = self.t_enc(input_ids=text_ids, attention_mask=text_mask, return_dict=True)
+        
+        a_pool = mean_pool(a_out.last_hidden_state, audio_mask)
+        t_pool = mean_pool(t_out.last_hidden_state, text_mask)
         v_pool = self.v_enc(video)
+        
+        a_pool = torch.nan_to_num(a_pool, nan=0.0, posinf=1e4, neginf=-1e4)
+        t_pool = torch.nan_to_num(t_pool, nan=0.0, posinf=1e4, neginf=-1e4)
+        v_pool = torch.nan_to_num(v_pool, nan=0.0, posinf=1e4, neginf=-1e4)
+        
+        a_logits = self.fc_audio(a_pool)
+        t_logits = self.fc_text(t_pool)
+        v_logits = self.fc_video(v_pool)
+        
+        fused = self.hbf(a_pool, t_pool, v_pool)
+        fused = torch.nan_to_num(fused, nan=0.0, posinf=1e4, neginf=-1e4)
+        fused_logits = self.fc(fused)
+        
+        return fused_logits, a_logits, t_logits, v_logits
 
-        fused = self.hbf(
-            a_pool,
-            t_pool,
-            v_pool
-        )
-
-        logits = self.fc(fused)
-
-        return logits
-
-# =========================================================
-# LOAD MODEL
-# =========================================================
+# Load model
 
-model = AVVideoModel(
-    num_classes=len(LABELS)
-).to(DEVICE)
+model = AVVideoModel(num_classes=len(LABELS)).to(DEVICE)
 
+# Download model from Hugging Face Model Hub
 try:
-
     model_path = hf_hub_download(
-        repo_id="ApurvaKondekar/emotion_model",
-        filename="model_weights.pth"
-    )
-
-    model.load_state_dict(
-        torch.load(model_path, map_location=DEVICE)
+        repo_id="ApurvaKondekar/emotion_model",  # CHANGE THIS
+        filename="model_weights.pth"  # YOUR FILE NAME
     )
-
+    model.load_state_dict(torch.load(model_path, map_location=DEVICE))
     model.eval()
-
-    print("Model loaded successfully")
-
+    print("✅ Model loaded from Hugging Face")
 except Exception as e:
-
-    print(f"Model loading failed: {e}")
-
-# =========================================================
-# VIDEO PROCESSING
-# =========================================================
-
-def extract_video_frames(
-    video_path,
-    max_frames=8,
-    resize=(224, 224)
-):
-
+    print(f"❌ Failed to load model: {e}")
+    # Alternative: Create a dummy model for testing
+    # Load trained weights (you'll need to upload this)
+    if os.path.exists("model_weights.pth"):
+        model.load_state_dict(torch.load("model_weights.pth", map_location=DEVICE))
+        model.eval()
+        print("✅ Model loaded successfully")
+    else:
+        print("⚠️ No model weights found. Using untrained model for demo.")
+
+def extract_video_frames(video_path, max_frames=8, resize=(224, 224)):
+    """Extract frames from video file"""
     cap = cv2.VideoCapture(video_path)
-
     if not cap.isOpened():
         return None
-
-    total_frames = int(
-        cap.get(cv2.CAP_PROP_FRAME_COUNT)
-    )
-
-    indices = np.linspace(
-        0,
-        total_frames - 1,
-        max_frames,
-        dtype=int
-    )
-
+    
     frames = []
-
-    for idx in indices:
-
-        cap.set(cv2.CAP_PROP_POS_FRAMES, idx)
-
+    while len(frames) < max_frames:
         ret, frame = cap.read()
-
         if not ret:
-            continue
-
-        frame = cv2.cvtColor(
-            frame,
-            cv2.COLOR_BGR2RGB
-        )
-
+            break
+        frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
         frame = cv2.resize(frame, resize)
-
         frames.append(frame)
-
+    
     cap.release()
-
+    
     if len(frames) == 0:
         return None
-
+    
+    # Pad if needed
     while len(frames) < max_frames:
         frames.append(frames[-1])
-
-    frames = np.array(frames[:max_frames])
-
+    
+    frames = np.array(frames[:max_frames], dtype=np.uint8)
     return frames
 
-# =========================================================
-# AUDIO EXTRACTION
-# =========================================================
-
 def extract_audio_from_video(video_path):
-
-    audio_path = tempfile.NamedTemporaryFile(
-        delete=False,
-        suffix=".wav"
-    ).name
-
-    command = [
-        "ffmpeg",
-        "-i", video_path,
-        "-vn",
-        "-acodec", "pcm_s16le",
-        "-ar", str(SAMPLE_RATE),
-        "-ac", "1",
-        "-y",
-        audio_path
-    ]
-
-    result = subprocess.run(
-        command,
-        stdout=subprocess.PIPE,
-        stderr=subprocess.PIPE
-    )
-
-    if result.returncode != 0:
-        error_msg = result.stderr.decode("utf-8", errors="replace")
-        print(f"[DEBUG] ffmpeg error: {error_msg}")
-        raise RuntimeError(
-            f"ffmpeg failed to extract audio: {error_msg[-200:]}"
-        )
-
-    if not os.path.exists(audio_path) or os.path.getsize(audio_path) == 0:
-        raise RuntimeError(
-            "ffmpeg produced no audio output — video may have no audio track"
-        )
-
-    return audio_path
-
-# =========================================================
-# WHISPER
-# =========================================================
-
-whisper_model = whisper.load_model("base")
+    """Extract audio from video file using ffmpeg"""
+    try:
+        audio_path = tempfile.NamedTemporaryFile(delete=False, suffix=".wav").name
+        # Use ffmpeg to extract audio
+        import subprocess
+        command = [
+            'ffmpeg', '-i', video_path,
+            '-vn',  # No video
+            '-acodec', 'pcm_s16le',  # Audio codec
+            '-ar', str(SAMPLE_RATE),  # Sample rate
+            '-ac', '1',  # Mono
+            '-y',  # Overwrite
+            audio_path
+        ]
+        subprocess.run(command, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL, check=True)
+        return audio_path
+    except Exception as e:
+        raise ValueError(f"Could not extract audio from video: {str(e)}")
 
 def transcribe_audio(audio_path):
+    """Transcribe audio using Whisper"""
+    try:
+        whisper_model = whisper.load_model("base")
+        result = whisper_model.transcribe(audio_path)
+        return result["text"].strip()
+    except Exception as e:
+        raise ValueError(f"Could not transcribe audio: {str(e)}")
 
-    result = whisper_model.transcribe(audio_path)
-
-    return result["text"].strip()
-
-# =========================================================
-# PREPROCESSING
-# =========================================================
-
-def preprocess_inputs(
-    audio_path,
-    text,
-    video_path
-):
-
-    wav, _ = librosa.load(
-        audio_path,
-        sr=SAMPLE_RATE
-    )
-
-    audio_inputs = processor(
-        wav,
-        sampling_rate=SAMPLE_RATE,
-        return_tensors="pt"
-    )
 
+def preprocess_inputs(audio_path, text, video_path):
+    """Preprocess all three modalities"""
+    
+    # Audio
+    wav, _ = librosa.load(audio_path, sr=SAMPLE_RATE)
+    audio_inputs = processor(wav, sampling_rate=SAMPLE_RATE, return_tensors="pt")
     audio_values = audio_inputs.input_values.to(DEVICE)
-
-    audio_mask = torch.ones_like(
-        audio_values
-    ).to(DEVICE)
-
-    text_clean = re.sub(
-        r"[^a-zA-Z0-9\s]",
-        "",
-        text.lower()
-    )
-
+    audio_mask = torch.ones_like(audio_values).to(DEVICE)
+    
+    # Text
+    text_clean = re.sub(r"[^a-zA-Z0-9\s]", "", text.lower())
     text_inputs = tokenizer(
         text_clean,
         truncation=True,
@@ -410,1244 +201,110 @@ def preprocess_inputs(
         max_length=TEXT_MAX_LEN,
         return_tensors="pt"
     )
-
     text_ids = text_inputs.input_ids.to(DEVICE)
-
     text_mask = text_inputs.attention_mask.to(DEVICE)
-
+    
+    # Video
     frames = extract_video_frames(video_path)
-
-    frames_tensor = (
-        torch.tensor(frames)
-        .permute(0, 3, 1, 2)
-        .float() / 255.0
-    )
-
-    frames_tensor = (
-        frames_tensor
-        .unsqueeze(0)
-        .permute(0, 2, 1, 3, 4)
-        .to(DEVICE)
-    )
-
-    return (
-        audio_values,
-        audio_mask,
-        text_ids,
-        text_mask,
-        frames_tensor
-    )
-
-# =========================================================
-# PREDICTION
-# =========================================================
+    if frames is None:
+        raise ValueError("Could not extract frames from video")
+    
+    frames_tensor = torch.tensor(frames).permute(0, 3, 1, 2).float() / 255.0
+    frames_tensor = frames_tensor.unsqueeze(0).permute(0, 2, 1, 3, 4).to(DEVICE)
+    
+    return audio_values, audio_mask, text_ids, text_mask, frames_tensor
 
 def predict_emotion(video_file):
-
+    """Main prediction function - takes only video input"""
+    
     if video_file is None:
-
-        return (
-            "⚠ NO INPUT DETECTED — UPLOAD A VIDEO FILE TO BEGIN ANALYSIS.",
-            None,
-            ""
-        )
-
+        return "Please provide a video file", None, ""
+    
     try:
-
-        # Handle both string paths and dict inputs from Gradio
-        if isinstance(video_file, dict):
-            video_path = video_file.get("video", video_file.get("name", None))
-        else:
-            video_path = video_file
-
-        if video_path is None or not os.path.exists(video_path):
-            return (
-                "⚠ VIDEO FILE NOT FOUND — The recorded/uploaded file could not be located.",
-                None,
-                ""
-            )
-
-        print(f"[DEBUG] Processing video: {video_path}")
-        print(f"[DEBUG] File exists: {os.path.exists(video_path)}")
-        print(f"[DEBUG] File size: {os.path.getsize(video_path)} bytes")
-
-        audio_path = extract_audio_from_video(
-            video_path
-        )
-
-        transcribed_text = transcribe_audio(
-            audio_path
-        )
-
-        (
-            audio,
-            audio_mask,
-            text_ids,
-            text_mask,
-            video
-        ) = preprocess_inputs(
-            audio_path,
-            transcribed_text,
-            video_path
+        # Extract audio from video
+        audio_path = extract_audio_from_video(video_file)
+        
+        # Transcribe audio
+        transcribed_text = transcribe_audio(audio_path)
+        
+        # Preprocess all modalities
+        audio, audio_mask, text_ids, text_mask, video = preprocess_inputs(
+            audio_path, transcribed_text, video_file
         )
-
+        
+        # Inference
         with torch.no_grad():
-
-            logits = model(
-                audio,
-                audio_mask,
-                text_ids,
-                text_mask,
-                video
+            fused_logits, a_logits, t_logits, v_logits = model(
+                audio, audio_mask, text_ids, text_mask, video
             )
-
-        probs = torch.softmax(
-            logits,
-            dim=1
-        )[0].cpu().numpy()
-
-        result = {
-            LABELS[i]: float(probs[i])
-            for i in range(len(LABELS))
-        }
-
-        predicted_emotion = LABELS[
-            probs.argmax()
-        ]
-
-        confidence = float(probs.max())
-
-        result_text = f"""
-        ## EMOTION DETECTED
         
-        ### `{predicted_emotion.upper()}`
+        # Get probabilities
+        fused_probs = torch.softmax(fused_logits, dim=1)[0].cpu().numpy()
         
-        **Neural Confidence:** `{confidence:.2%}`
-        """
-
+        # Format results
+        result = {LABELS[i]: float(fused_probs[i]) for i in range(len(LABELS))}
+        
+        predicted_emotion = LABELS[fused_probs.argmax()]
+        confidence = float(fused_probs.max())
+        
+        result_text = f"🎯 **Predicted Emotion: {predicted_emotion.upper()}**\n\n**Confidence: {confidence:.2%}**"
+        
+        # Clean up temporary audio file
         if os.path.exists(audio_path):
             os.remove(audio_path)
-
-        return (
-            result_text,
-            result,
-            transcribed_text
-        )
-
+        
+        return result_text, result, transcribed_text
+        
     except Exception as e:
+        return f"Error: {str(e)}", None, ""
 
-        return (
-            f"⚠ SYSTEM ERROR: {str(e)}",
-            None,
-            ""
-        )
-
-# =========================================================
-# FUTURISTIC CYBERPUNK CSS
-# =========================================================
-
-custom_css = """
-/* =====================================================
-   IMPORT FONTS
-===================================================== */
-@import url('https://fonts.googleapis.com/css2?family=Orbitron:wght@400;600;700;900&family=Share+Tech+Mono&family=Rajdhani:wght@300;400;600;700&display=swap');
-
-/* =====================================================
-   CSS VARIABLES
-===================================================== */
-:root {
-    --neon-cyan:    #00f5ff;
-    --neon-green:   #00ff88;
-    --neon-pink:    #ff006e;
-    --neon-purple:  #bf00ff;
-    --neon-orange:  #ff6b35;
-    --dark-bg:      #060e1a;
-    --dark-panel:   #0a1628;
-    --dark-card:    #0d1e35;
-    --border-glow:  rgba(0, 245, 255, 0.35);
-    --text-primary: #cff4ff;
-    --text-bright:  #ffffff;
-    --text-dim:     #7eb8cc;
-    --text-muted:   #5a9ab0;
-    --grid-color:   rgba(0, 245, 255, 0.05);
-}
-
-/* =====================================================
-   GLOBAL RESET & BASE
-===================================================== */
-* {
-    box-sizing: border-box;
-}
-
-body, .gradio-container {
-    background-color: var(--dark-bg) !important;
-    font-family: 'Rajdhani', sans-serif !important;
-    color: var(--text-primary) !important;
-    min-height: 100vh;
-    position: relative;
-    overflow-x: hidden;
-}
-
-/* Animated grid background */
-body::before {
-    content: '';
-    position: fixed;
-    inset: 0;
-    background-image:
-        linear-gradient(var(--grid-color) 1px, transparent 1px),
-        linear-gradient(90deg, var(--grid-color) 1px, transparent 1px);
-    background-size: 40px 40px;
-    z-index: 0;
-    pointer-events: none;
-    animation: gridDrift 20s linear infinite;
-}
-
-/* Scanline overlay */
-body::after {
-    content: '';
-    position: fixed;
-    inset: 0;
-    background: repeating-linear-gradient(
-        0deg,
-        transparent,
-        transparent 2px,
-        rgba(0, 245, 255, 0.015) 2px,
-        rgba(0, 245, 255, 0.015) 4px
-    );
-    pointer-events: none;
-    z-index: 1;
-    animation: scanlines 8s linear infinite;
-}
-
-@keyframes gridDrift {
-    0%   { background-position: 0 0; }
-    100% { background-position: 40px 40px; }
-}
-
-@keyframes scanlines {
-    0%   { background-position: 0 0; }
-    100% { background-position: 0 100px; }
-}
-
-/* =====================================================
-   CONTAINER
-===================================================== */
-.gradio-container {
-    max-width: 1400px !important;
-    margin: auto !important;
-    padding: 30px 24px !important;
-    position: relative;
-    z-index: 2;
-}
-
-/* =====================================================
-   HEADER
-===================================================== */
-.cyber-header {
-    text-align: center;
-    padding: 40px 20px 20px;
-    position: relative;
-}
-
-.cyber-title {
-    font-family: 'Orbitron', monospace !important;
-    font-size: clamp(28px, 5vw, 58px) !important;
-    font-weight: 900 !important;
-    letter-spacing: 6px !important;
-    text-transform: uppercase !important;
-    color: transparent !important;
-    background: linear-gradient(
-        90deg,
-        var(--neon-cyan) 0%,
-        #ffffff 40%,
-        var(--neon-purple) 70%,
-        var(--neon-cyan) 100%
-    ) !important;
-    background-size: 200% auto !important;
-    -webkit-background-clip: text !important;
-    background-clip: text !important;
-    animation: titleShimmer 4s linear infinite, fadeSlideDown 0.8s ease both;
-    text-shadow: none !important;
-    position: relative;
-}
-
-.cyber-title::after {
-    content: attr(data-text);
-    position: absolute;
-    left: 0; right: 0;
-    top: 0;
-    color: var(--neon-cyan);
-    filter: blur(18px);
-    opacity: 0.35;
-    animation: titlePulse 3s ease-in-out infinite;
-    z-index: -1;
-}
-
-@keyframes titleShimmer {
-    0%   { background-position: 0% center; }
-    100% { background-position: 200% center; }
-}
-
-@keyframes titlePulse {
-    0%, 100% { opacity: 0.25; }
-    50%       { opacity: 0.5; }
-}
-
-@keyframes fadeSlideDown {
-    from { opacity: 0; transform: translateY(-24px); }
-    to   { opacity: 1; transform: translateY(0); }
-}
-
-.cyber-subtitle {
-    font-family: 'Share Tech Mono', monospace !important;
-    font-size: 14px !important;
-    color: var(--text-dim) !important;
-    letter-spacing: 3px !important;
-    text-transform: uppercase !important;
-    margin-top: 10px !important;
-    animation: fadeSlideDown 1s ease 0.3s both;
-}
-
-/* Decorative line under title */
-.cyber-divider {
-    display: flex;
-    align-items: center;
-    gap: 12px;
-    margin: 20px auto;
-    max-width: 600px;
-    animation: fadeSlideDown 1s ease 0.5s both;
-}
-
-.cyber-divider::before,
-.cyber-divider::after {
-    content: '';
-    flex: 1;
-    height: 1px;
-    background: linear-gradient(90deg, transparent, var(--neon-cyan), transparent);
-}
-
-.cyber-divider-dot {
-    width: 6px; height: 6px;
-    background: var(--neon-cyan);
-    border-radius: 50%;
-    box-shadow: 0 0 10px var(--neon-cyan);
-    animation: dotPulse 2s ease-in-out infinite;
-}
-
-@keyframes dotPulse {
-    0%, 100% { transform: scale(1); opacity: 1; }
-    50%       { transform: scale(1.6); opacity: 0.6; }
-}
-
-/* =====================================================
-   STATUS BADGE
-===================================================== */
-.status-bar {
-    display: flex;
-    justify-content: center;
-    margin: 10px 0 24px;
-    animation: fadeSlideDown 1s ease 0.6s both;
-}
-
-.status-badge {
-    font-family: 'Share Tech Mono', monospace;
-    font-size: 12px;
-    letter-spacing: 2px;
-    padding: 6px 20px;
-    border: 1px solid var(--neon-green);
-    color: var(--neon-green);
-    background: rgba(0, 255, 136, 0.06);
-    border-radius: 2px;
-    position: relative;
-    overflow: hidden;
-    text-transform: uppercase;
-}
-
-.status-badge::before {
-    content: '';
-    position: absolute;
-    top: 0; left: -100%;
-    width: 100%; height: 100%;
-    background: linear-gradient(90deg, transparent, rgba(0,255,136,0.15), transparent);
-    animation: scanSweep 3s linear infinite;
-}
-
-@keyframes scanSweep {
-    0%   { left: -100%; }
-    100% { left: 100%; }
-}
-
-/* =====================================================
-   INFO CARDS (Modalities / Models)
-===================================================== */
-.info-grid {
-    display: grid;
-    grid-template-columns: 1fr 1fr;
-    gap: 16px;
-    margin-bottom: 24px;
-    animation: fadeSlideDown 1s ease 0.7s both;
-}
-
-.cyber-card {
-    background: #0d1e35 !important;
-    border: 1px solid rgba(0, 245, 255, 0.35) !important;
-    border-radius: 6px !important;
-    padding: 0 !important;
-    position: relative;
-    overflow: hidden;
-    transition: border-color 0.3s, box-shadow 0.3s;
-}
-
-.cyber-card-purple {
-    border-color: rgba(191, 0, 255, 0.35) !important;
-}
-
-/* Glowing left bar */
-.cyber-card::before {
-    content: '';
-    position: absolute;
-    top: 0; left: 0;
-    width: 3px; height: 100%;
-    background: linear-gradient(180deg, #00f5ff, #bf00ff);
-    box-shadow: 0 0 14px #00f5ff;
-    z-index: 0;
-}
-
-.cyber-card-purple::before {
-    background: linear-gradient(180deg, #bf00ff, #00f5ff);
-    box-shadow: 0 0 14px #bf00ff;
-}
-
-/* Corner glow — behind text */
-.cyber-card::after {
-    content: '';
-    position: absolute;
-    top: 0; right: 0;
-    width: 80px; height: 80px;
-    background: radial-gradient(circle, rgba(0,245,255,0.07) 0%, transparent 70%);
-    z-index: 0;
-}
-
-/* Content wrapper sits above pseudo-elements */
-.card-inner {
-    position: relative;
-    z-index: 2;
-    padding: 22px 22px 22px 28px;
-}
-
-.card-label {
-    font-family: 'Orbitron', monospace;
-    font-size: 10px;
-    font-weight: 700;
-    letter-spacing: 3px;
-    text-transform: uppercase;
-    margin-bottom: 16px;
-}
-
-.card-row {
-    font-family: 'Share Tech Mono', monospace;
-    font-size: 14px;
-    color: #ffffff !important;
-    line-height: 2.2;
-    display: flex;
-    align-items: center;
-    gap: 10px;
-    opacity: 1 !important;
-}
-
-.card-dot {
-    font-size: 12px;
-    flex-shrink: 0;
-}
-
-.cyber-card:hover {
-    border-color: rgba(0, 245, 255, 0.6) !important;
-    box-shadow: 0 0 35px rgba(0, 245, 255, 0.12) !important;
-}
-
-.cyber-card-purple:hover {
-    border-color: rgba(191, 0, 255, 0.6) !important;
-    box-shadow: 0 0 35px rgba(191, 0, 255, 0.12) !important;
-}
-
-/* =====================================================
-   SYSTEM ARCHITECTURE STYLES
-===================================================== */
-.arch-grid {
-    display: grid;
-    grid-template-columns: 3fr 2fr;
-    gap: 24px;
-    padding: 20px 4px 8px;
-}
-
-.arch-section {
-    display: flex;
-    flex-direction: column;
-    gap: 14px;
-}
-
-.arch-title {
-    font-family: 'Orbitron', monospace;
-    font-size: 11px;
-    font-weight: 700;
-    letter-spacing: 3px;
-    color: #00f5ff;
-    text-transform: uppercase;
-    border-bottom: 1px solid rgba(0,245,255,0.2);
-    padding-bottom: 8px;
-    margin-bottom: 4px;
-}
-
-.arch-desc {
-    font-family: 'Rajdhani', sans-serif;
-    font-size: 15px;
-    color: #ffffff !important;
-    line-height: 1.8;
-    margin: 0 0 8px;
-}
-
-.arch-row {
-    display: flex;
-    align-items: flex-start;
-    gap: 14px;
-}
-
-.arch-tag {
-    font-family: 'Orbitron', monospace;
-    font-size: 9px;
-    font-weight: 700;
-    letter-spacing: 1.5px;
-    border: 1px solid;
-    border-radius: 2px;
-    padding: 4px 8px;
-    white-space: nowrap;
-    flex-shrink: 0;
-    margin-top: 2px;
-}
-
-.arch-detail {
-    font-family: 'Share Tech Mono', monospace;
-    font-size: 13px;
-    color: #ffffff !important;
-    line-height: 1.8;
-    opacity: 1 !important;
-}
-
-.emotion-chips {
-    display: grid;
-    grid-template-columns: 1fr 1fr;
-    gap: 10px;
-    margin-top: 4px;
-}
-
-.emotion-chip {
-    font-family: 'Orbitron', monospace;
-    font-size: 11px;
-    font-weight: 700;
-    letter-spacing: 2px;
-    border: 1px solid;
-    border-radius: 4px;
-    padding: 12px 10px;
-    text-align: center;
-    background: rgba(255,255,255,0.02);
-    transition: all 0.3s;
-}
-
-.emotion-chip:hover {
-    background: rgba(255,255,255,0.06);
-    transform: translateY(-2px);
-}
-
-/* =====================================================
-   HIDE GRADIO "USE API" / BUILT WITH FOOTER
-===================================================== */
-.built-with,
-a[href*="gradio"],
-footer.svelte-1ax1toq,
-.svelte-1ax1toq,
-div[class*="built"],
-.show-api,
-button[title*="API"],
-.api-btn {
-    display: none !important;
-}
-
-/* =====================================================
-   GR PANELS (override Gradio)
-===================================================== */
-.gr-group,
-.gr-box,
-.gr-panel,
-.gr-form,
-div[class*="panel"],
-div[class*="block"] {
-    background: #0a1628 !important;
-    border: 1px solid rgba(0, 245, 255, 0.2) !important;
-    border-radius: 4px !important;
-    box-shadow: none !important;
-    backdrop-filter: none !important;
-}
-
-/* =====================================================
-   MARKDOWN HEADINGS
-===================================================== */
-h1, h2, h3, h4 {
-    font-family: 'Orbitron', monospace !important;
-    font-weight: 700 !important;
-    letter-spacing: 2px !important;
-    text-transform: uppercase !important;
-}
-
-h2 {
-    font-size: 16px !important;
-    color: #00f5ff !important;
-    border-bottom: 1px solid rgba(0,245,255,0.3) !important;
-    padding-bottom: 8px !important;
-    margin-bottom: 16px !important;
-    text-shadow: 0 0 12px rgba(0,245,255,0.5) !important;
-}
-
-h3 {
-    font-size: 15px !important;
-    color: #ffffff !important;
-}
-
-h4 {
-    font-size: 13px !important;
-    color: #7eb8cc !important;
-}
-
-/* =====================================================
-   LABELS / TEXT
-===================================================== */
-label, .gr-label span, p, li {
-    font-family: 'Rajdhani', sans-serif !important;
-    color: #ffffff !important;
-    font-size: 15px !important;
-    font-weight: 600 !important;
-    letter-spacing: 1px !important;
-}
-
-/* Gradio component labels */
-.svelte-1ipelgc, span.svelte-1ipelgc,
-div[class*="label"] > span,
-.block > label > span {
-    color: var(--neon-cyan) !important;
-    font-family: 'Share Tech Mono', monospace !important;
-    font-size: 12px !important;
-    letter-spacing: 2px !important;
-    text-transform: uppercase !important;
-    opacity: 1 !important;
-}
-
-/* =====================================================
-   ANALYZE BUTTON
-===================================================== */
-.gr-button,
-button[class*="primary"],
-button {
-    font-family: 'Orbitron', monospace !important;
-    font-size: 13px !important;
-    font-weight: 700 !important;
-    letter-spacing: 3px !important;
-    text-transform: uppercase !important;
-    background: transparent !important;
-    border: 1px solid var(--neon-cyan) !important;
-    color: var(--neon-cyan) !important;
-    border-radius: 3px !important;
-    padding: 14px 32px !important;
-    position: relative !important;
-    overflow: hidden !important;
-    transition: all 0.3s ease !important;
-    cursor: pointer !important;
-}
-
-.gr-button::before,
-button::before {
-    content: '' !important;
-    position: absolute !important;
-    top: 0; left: -100% !important;
-    width: 100%; height: 100% !important;
-    background: linear-gradient(
-        90deg,
-        transparent,
-        rgba(0, 245, 255, 0.2),
-        transparent
-    ) !important;
-    transition: left 0.5s ease !important;
-}
-
-.gr-button:hover::before,
-button:hover::before {
-    left: 100% !important;
-}
-
-.gr-button:hover,
-button:hover {
-    background: rgba(0, 245, 255, 0.1) !important;
-    box-shadow:
-        0 0 20px rgba(0, 245, 255, 0.4),
-        0 0 60px rgba(0, 245, 255, 0.15),
-        inset 0 0 20px rgba(0, 245, 255, 0.05) !important;
-    transform: translateY(-2px) !important;
-}
-
-.gr-button:active,
-button:active {
-    transform: translateY(0px) !important;
-}
-
-/* =====================================================
-   INPUT / TEXTAREA
-===================================================== */
-textarea,
-input[type="text"],
-input {
-    font-family: 'Share Tech Mono', monospace !important;
-    font-size: 13px !important;
-    background: #071020 !important;
-    border: 1px solid rgba(0, 245, 255, 0.25) !important;
-    border-radius: 3px !important;
-    color: #cff4ff !important;
-    transition: all 0.3s !important;
-    letter-spacing: 0.5px !important;
-}
-
-textarea:focus,
-input:focus {
-    border-color: var(--neon-cyan) !important;
-    box-shadow: 0 0 15px rgba(0, 245, 255, 0.25) !important;
-    outline: none !important;
-}
-
-/* placeholder */
-textarea::placeholder,
-input::placeholder {
-    color: rgba(127, 184, 204, 0.45) !important;
-}
-
-/* =====================================================
-   VIDEO COMPONENT
-===================================================== */
-video {
-    border-radius: 3px !important;
-    border: 1px solid rgba(0, 245, 255, 0.2) !important;
-    box-shadow:
-        0 0 30px rgba(0, 245, 255, 0.1),
-        inset 0 0 30px rgba(0, 0, 0, 0.5) !important;
-}
-
-/* Upload zone */
-.upload-zone,
-div[class*="upload"] {
-    border: 1px dashed rgba(0, 245, 255, 0.3) !important;
-    background: rgba(0, 245, 255, 0.03) !important;
-    border-radius: 4px !important;
-    transition: all 0.3s !important;
-}
-
-.upload-zone:hover,
-div[class*="upload"]:hover {
-    border-color: var(--neon-cyan) !important;
-    background: rgba(0, 245, 255, 0.07) !important;
-}
-
-/* =====================================================
-   REMOVE VIDEO EDIT CONTROLS
-===================================================== */
-
-/* remove trim button */
-button[aria-label*="Trim"],
-button[title*="Trim"],
-button[aria-label*="trim"],
-button[title*="trim"] {
-    display: none !important;
-}
-
-/* remove reset button */
-button[aria-label*="Reset"],
-button[title*="Reset"],
-button[aria-label*="reset"],
-button[title*="reset"] {
-    display: none !important;
-}
-
-/* remove edit tools section */
-div[data-testid="video"] .controls,
-div[data-testid="video"] [class*="controls"],
-div[data-testid="video"] [class*="edit"] {
-    display: none !important;
-}
-
-/* =====================================================
-   LABEL (EMOTION PROBABILITIES)
-===================================================== */
-.gr-label,
-div[class*="label"] {
-    background: var(--dark-panel) !important;
-    border: 1px solid rgba(0, 245, 255, 0.12) !important;
-    border-radius: 4px !important;
-    padding: 12px !important;
-}
-
-/* Label bars */
-div[class*="confidence"] span,
-div[class*="bar"],
-.label-bar,
-[class*="Confidence"] {
-    background: linear-gradient(
-        90deg,
-        rgba(0, 245, 255, 0.15),
-        rgba(0, 245, 255, 0.05)
-    ) !important;
-    border-left: 2px solid var(--neon-cyan) !important;
-    border-radius: 0 !important;
-}
-
-/* =====================================================
-   ACCORDION
-===================================================== */
-.gr-accordion,
-details,
-summary {
-    background: #0d1e35 !important;
-    border: 1px solid rgba(0,245,255,0.2) !important;
-    border-radius: 4px !important;
-    font-family: 'Orbitron', monospace !important;
-    font-size: 12px !important;
-    letter-spacing: 2px !important;
-    color: #7eb8cc !important;
-    text-transform: uppercase !important;
-}
-
-details summary {
-    padding: 14px 20px !important;
-    cursor: pointer !important;
-    transition: color 0.3s !important;
-    color: #a0d4e8 !important;
-}
-
-details summary:hover {
-    color: var(--neon-cyan) !important;
-}
-
-details[open] summary {
-    color: var(--neon-cyan) !important;
-    border-bottom: 1px solid rgba(0,245,255,0.2) !important;
-}
-
-/* =====================================================
-   RESULT TEXT (Markdown output)
-===================================================== */
-.result-markdown code,
-code {
-    font-family: 'Share Tech Mono', monospace !important;
-    background: rgba(0, 245, 255, 0.08) !important;
-    border: 1px solid rgba(0, 245, 255, 0.2) !important;
-    border-radius: 2px !important;
-    color: var(--neon-cyan) !important;
-    padding: 2px 8px !important;
-    font-size: 15px !important;
-}
-
-/* =====================================================
-   SEPARATOR / HR
-===================================================== */
-hr {
-    border: none !important;
-    height: 1px !important;
-    background: linear-gradient(
-        90deg,
-        transparent,
-        rgba(0, 245, 255, 0.3),
-        rgba(191, 0, 255, 0.3),
-        transparent
-    ) !important;
-    margin: 24px 0 !important;
-}
-
-/* =====================================================
-   FOOTER
-===================================================== */
-.cyber-footer {
-    text-align: center;
-    font-family: 'Share Tech Mono', monospace;
-    font-size: 11px;
-    letter-spacing: 3px;
-    color: var(--text-dim);
-    margin-top: 40px;
-    padding: 20px;
-    text-transform: uppercase;
-    border-top: 1px solid rgba(0,245,255,0.08);
-    position: relative;
-}
-
-.cyber-footer::before {
-    content: '◈ ◈ ◈';
-    display: block;
-    color: rgba(0,245,255,0.2);
-    margin-bottom: 10px;
-    letter-spacing: 6px;
-    font-size: 8px;
-    animation: dotPulse 3s ease-in-out infinite;
-}
-
-/* =====================================================
-   CORNER DECORATION (applied to main columns)
-===================================================== */
-.corner-decor {
-    position: relative;
-}
-
-.corner-decor::before,
-.corner-decor::after {
-    content: '';
-    position: absolute;
-    width: 12px; height: 12px;
-    border-color: var(--neon-cyan);
-    border-style: solid;
-    opacity: 0.5;
-}
-
-.corner-decor::before {
-    top: -1px; left: -1px;
-    border-width: 2px 0 0 2px;
-}
-
-.corner-decor::after {
-    bottom: -1px; right: -1px;
-    border-width: 0 2px 2px 0;
-}
-
-/* =====================================================
-   ANIMATED PARTICLES (pure CSS)
-===================================================== */
-.particle-field {
-    position: fixed;
-    inset: 0;
-    pointer-events: none;
-    z-index: 0;
-    overflow: hidden;
-}
-
-.particle {
-    position: absolute;
-    width: 2px; height: 2px;
-    background: var(--neon-cyan);
-    border-radius: 50%;
-    opacity: 0;
-    animation: floatParticle linear infinite;
-}
-
-.particle:nth-child(1)  { left: 10%; animation-duration: 12s; animation-delay: 0s;   width: 1px; height: 1px; }
-.particle:nth-child(2)  { left: 20%; animation-duration: 18s; animation-delay: 2s;   background: var(--neon-purple); }
-.particle:nth-child(3)  { left: 35%; animation-duration: 14s; animation-delay: 4s;   }
-.particle:nth-child(4)  { left: 50%; animation-duration: 20s; animation-delay: 1s;   background: var(--neon-green); width: 1px; height: 1px; }
-.particle:nth-child(5)  { left: 65%; animation-duration: 16s; animation-delay: 6s;   }
-.particle:nth-child(6)  { left: 78%; animation-duration: 13s; animation-delay: 3s;   background: var(--neon-pink); }
-.particle:nth-child(7)  { left: 88%; animation-duration: 19s; animation-delay: 5s;   width: 1px; height: 1px; }
-.particle:nth-child(8)  { left: 45%; animation-duration: 15s; animation-delay: 7s;   background: var(--neon-purple); }
-.particle:nth-child(9)  { left: 55%; animation-duration: 17s; animation-delay: 0.5s; }
-.particle:nth-child(10) { left: 72%; animation-duration: 11s; animation-delay: 8s;   background: var(--neon-green); width: 1px; height: 1px; }
-.particle:nth-child(11) { left: 5%;  animation-duration: 22s; animation-delay: 2.5s; background: var(--neon-cyan); }
-.particle:nth-child(12) { left: 92%; animation-duration: 14s; animation-delay: 9s;   background: var(--neon-pink); width: 1px; height: 1px; }
-
-@keyframes floatParticle {
-    0%   { bottom: -10px; opacity: 0; transform: translateX(0); }
-    10%  { opacity: 0.6; }
-    90%  { opacity: 0.3; }
-    100% { bottom: 105vh; opacity: 0; transform: translateX(30px); }
-}
-
-/* =====================================================
-   SCROLLBAR
-===================================================== */
-::-webkit-scrollbar { width: 6px; }
-::-webkit-scrollbar-track { background: var(--dark-bg); }
-::-webkit-scrollbar-thumb {
-    background: linear-gradient(180deg, var(--neon-cyan), var(--neon-purple));
-    border-radius: 3px;
-}
-
-/* =====================================================
-   RESPONSIVE TWEAKS
-===================================================== */
-@media (max-width: 768px) {
-    .info-grid { grid-template-columns: 1fr; }
-    .cyber-title { font-size: 26px !important; letter-spacing: 3px !important; }
-}
-
-/* REMOVE GRADIO FOOTER + SETTINGS */
-
-footer,
-.settings,
-button[title="Settings"],
-button[aria-label="Settings"],
-.gradio-footer,
-div[class*="settings"],
-div[class*="footer"] {
-    display: none !important;
-}
-
-/* =====================================================
-   CYBERPUNK VIDEO BUTTON FIX
-===================================================== */
-
-/* upload / record buttons */
-div[data-testid="video"] button * {
-    color: #ffffff !important;
-    fill: #ffffff !important;
-    stroke: #ffffff !important;
-    opacity: 1 !important;
-    visibility: visible !important;
-    font-size: 16px !important;
-    font-weight: 700 !important;
-}
-
-
-/* =====================================================
-   CYBERPUNK VIDEO BUTTON FIX
-===================================================== */
-
-/* upload / record buttons */
-div[data-testid="video"] button {
-    background: #071020 !important;
-    border: 2px solid #00f5ff !important;
-    border-radius: 6px !important;
-
-    min-width: 130px !important;
-    min-height: 40px !important;
-
-    display: flex !important;
-    align-items: center !important;
-    justify-content: center !important;
-    gap: 10px !important;
-
-    padding: 10px 18px !important;
-
-    color: #ffffff !important;
-    opacity: 1 !important;
-
-    box-shadow: 0 0 12px rgba(0,245,255,0.25) !important;
-}
-
-/* icon size */
-div[data-testid="video"] button svg {
-    width: 25px !important;
-    height: 25px !important;
-
-    stroke: #ffffff !important;
-    fill: #ffffff !important;
-
-    opacity: 1 !important;
-}
-
-/* text beside icon */
-div[data-testid="video"] button span {
-    color: #ffffff !important;
-
-    font-size: 16px !important;
-    font-weight: 700 !important;
-
-    letter-spacing: 1px !important;
-
-    opacity: 1 !important;
-    visibility: visible !important;
-
-    display: inline !important;
-}
-
-/* remove shiny animation overlay */
-div[data-testid="video"] button::before {
-    display: none !important;
-}
-
-/* black labels */
-.black-label label,
-.black-label span {
-    color: black !important;
-} 
-"""
-
-# =========================================================
-# UI
-# =========================================================
-
-with gr.Blocks(
-    title="Multimodal Emotion Recognition",
-    theme=gr.themes.Base(
-        primary_hue="cyan",
-        neutral_hue="slate",
-        font=[gr.themes.GoogleFont("Rajdhani"), "sans-serif"],
-    ),
-    css=custom_css
-) as demo:
-
-    # ── Floating particles ──────────────────────────────
-    gr.HTML("""
-    <div class="particle-field">
-        <div class="particle"></div>
-        <div class="particle"></div>
-        <div class="particle"></div>
-        <div class="particle"></div>
-        <div class="particle"></div>
-        <div class="particle"></div>
-        <div class="particle"></div>
-        <div class="particle"></div>
-        <div class="particle"></div>
-        <div class="particle"></div>
-        <div class="particle"></div>
-        <div class="particle"></div>
-    </div>
-    """)
-
-    # ── Header ──────────────────────────────────────────
-    gr.HTML("""
-    <div class="cyber-header">
-        <div class="cyber-title" data-text="MULTIMODAL EMOTION RECOGNITION">
-            MULTIMODAL EMOTION RECOGNITION
-        </div>
-        <div class="cyber-divider">
-            <div class="cyber-divider-dot"></div>
-            <div class="cyber-divider-dot" style="animation-delay:0.4s"></div>
-            <div class="cyber-divider-dot" style="animation-delay:0.8s"></div>
-        </div>
-    </div>
-    """)
-
-    # ── Info Cards ──────────────────────────────────────
-    gr.HTML("""
-    <div class="info-grid">
-        <div class="cyber-card">
-            <div class="card-inner">
-                <div class="card-label" style="color:#00f5ff;">MODALITIES</div>
-                <div class="card-row"><span class="card-dot" style="color:#00f5ff;">◈</span>Audio Waveform Analysis</div>
-                <div class="card-row"><span class="card-dot" style="color:#00f5ff;">◈</span>Speech Transcription</div>
-                <div class="card-row"><span class="card-dot" style="color:#00f5ff;">◈</span>Facial Expression Mapping</div>
-            </div>
-        </div>
-        <div class="cyber-card cyber-card-purple">
-            <div class="card-inner">
-                <div class="card-label" style="color:#bf00ff;">NEURAL MODELS</div>
-                <div class="card-row"><span class="card-dot" style="color:#bf00ff;">◈</span>Wav2Vec2 · BERT · ResNet18</div>
-                <div class="card-row"><span class="card-dot" style="color:#bf00ff;">◈</span>Whisper ASR</div>
-                <div class="card-row"><span class="card-dot" style="color:#bf00ff;">◈</span>Hierarchical Bottleneck Fusion (HBF)</div>
-            </div>
-        </div>
-    </div>
-    """)
-
-    gr.HTML('<hr/>')
-
-    # ── Main Analysis Section ───────────────────────────
-    with gr.Row(equal_height=True):
-
-        # LEFT: Upload
-        with gr.Column(scale=1):
-
-            gr.Markdown("## INPUT STREAM")
-
-            video_input = gr.Video(
-                label="Video Input",
-                height=360,
-                sources=["upload", "webcam"],
-                elem_classes="black-label"
-            )
-
-            predict_btn = gr.Button(
-                "ANALYZE EMOTION",
-                variant="primary",
-                size="lg"
-            )
-
-        # RIGHT: Results
-        with gr.Column(scale=1):
-
-            gr.Markdown("## ANALYSIS OUTPUT")
-
-            result_text = gr.Markdown(
-                value="""
-> `AWAITING INPUT` — Upload a video file and trigger analysis to begin neural processing.
-                """
-            )
-
-            result_output = gr.Label(
-                label="Emotion Probability Distribution",
-                num_top_classes=4,
-                 elem_classes="black-label"
-            )
-
-            transcription_output = gr.Textbox(
-                label="Transcribed Speech",
-                lines=5,
-                interactive=False,
-                placeholder="Speech transcription will appear here after analysis..."
-            )
-
-    gr.HTML('<hr/>')
-
-    # ── About Accordion ─────────────────────────────────
-    with gr.Accordion("SYSTEM ARCHITECTURE", open=False):
-
-        gr.HTML("""
-        <div class="arch-grid">
-            <div class="arch-section">
-                <div class="arch-title">NEURAL PIPELINE</div>
-                <p class="arch-desc">Three independent encoders fused through a Hybrid Bimodal Fusion (HBF) block across 6 iterative layers.</p>
-                <div class="arch-row">
-                    <div class="arch-tag" style="border-color:#00f5ff;color:#00f5ff;">WAV2VEC2</div>
-                    <div class="arch-detail">Extracts deep acoustic features from raw audio waveform</div>
-                </div>
-                <div class="arch-row">
-                    <div class="arch-tag" style="border-color:#bf00ff;color:#bf00ff;">BERT</div>
-                    <div class="arch-detail">Understands semantic and emotional tone from transcribed speech</div>
-                </div>
-                <div class="arch-row">
-                    <div class="arch-tag" style="border-color:#00ff88;color:#00ff88;">RESNET18</div>
-                    <div class="arch-detail">Captures facial micro-expressions across sampled video frames</div>
-                </div>
-                <div class="arch-row">
-                    <div class="arch-tag" style="border-color:#ff6b35;color:#ff6b35;">HBF</div>
-                    <div class="arch-detail">Iteratively aligns all three modalities over 6 fusion layers</div>
-                </div>
-            </div>
-            <div class="arch-section">
-                <div class="arch-title">DETECTABLE EMOTIONS</div>
-                <div class="emotion-chips">
-                    <div class="emotion-chip" style="border-color:#ff4444;color:#ff4444;box-shadow:0 0 12px rgba(255,68,68,0.2);">⚡ ANGRY</div>
-                    <div class="emotion-chip" style="border-color:#ffcc00;color:#ffcc00;box-shadow:0 0 12px rgba(255,204,0,0.2);">◈ HAPPY</div>
-                    <div class="emotion-chip" style="border-color:#00f5ff;color:#00f5ff;box-shadow:0 0 12px rgba(0,245,255,0.2);">◎ NEUTRAL</div>
-                    <div class="emotion-chip" style="border-color:#7b9fff;color:#7b9fff;box-shadow:0 0 12px rgba(123,159,255,0.2);">◇ SAD</div>
-                </div>
-            </div>
-        </div>
-        """)
-
-    # ── Footer ──────────────────────────────────────────
-    gr.HTML("""
-    <div class="cyber-footer">
-        Built on PyTorch &nbsp;·&nbsp; HuggingFace Transformers &nbsp;·&nbsp; OpenAI Whisper &nbsp;·&nbsp; Gradio
-    </div>
-    """)
+# Gradio Interface
+with gr.Blocks(title="Multimodal Emotion Recognition", theme=gr.themes.Soft()) as demo:
+    gr.Markdown(
+        """
+        # 🎭 Multimodal Emotion Recognition
+        
+        This system predicts emotions from video by automatically extracting and analyzing:
+        - 🎤 **Audio** (extracted from video)
+        - 📝 **Text** (transcribed from audio using Whisper)
+        - 🎥 **Video** (visual frames)
+        
+        ### How to use:
+        1. Upload a video file (MP4, AVI, MOV, etc.)
+        2. Click "Predict Emotion"
+        3. The system will automatically extract audio, transcribe speech, and analyze all modalities
+        
+        The model will provide emotion predictions based on all three inputs.
+        """
+    )
 
-    # ── Button Action ────────────────────────────────────
+    
+    with gr.Row():
+        with gr.Column():
+            video_input = gr.Video(label="🎥 Video Input")
+            predict_btn = gr.Button("🚀 Predict Emotion", variant="primary", size="lg")
+        
+        with gr.Column():
+            result_text = gr.Markdown(label="Result")
+            result_output = gr.Label(label="📊 Prediction Results", num_top_classes=4)
+            transcription_output = gr.Textbox(label="📝 Transcribed Text", lines=3, interactive=False)
+    
     predict_btn.click(
         fn=predict_emotion,
         inputs=[video_input],
-        outputs=[
-            result_text,
-            result_output,
-            transcription_output
-        ],
-        show_progress=True
+        outputs=[result_text, result_output, transcription_output]
     )
 
-# =========================================================
-# LAUNCH
-# =========================================================
+    
+    gr.Markdown(
+        """
+        ---
+        ### 📌 Notes:
+        - Supported emotions: **Angry, Happy, Neutral, Sad**
+        - Model uses Wav2Vec2 (audio), BERT (text), and ResNet18 (video)
+        - Best results with clear audio, accurate transcripts, and visible faces
+        """
+    )
 
 if __name__ == "__main__":
-
-    demo.queue()
-
     demo.launch()