Update app.py

app.py

@@ -1,4 +1,5 @@
-# integrated_voice_chat.py
+# ======================= app.py =======================
+
 import os
 import re
 import random
@@ -6,111 +7,132 @@ import tempfile
 import warnings
 warnings.filterwarnings("ignore")
 
-import gradio as gr
 import numpy as np
 import pandas as pd
 import soundfile as sf
 import librosa
 import joblib
-import requests
-
-# ML libs (optional)
-try:
-    import tensorflow as tf
-    from tensorflow.keras import layers, models
-    TF_AVAILABLE = True
-except Exception:
-    TF_AVAILABLE = False
-
+import gradio as gr
+from gtts import gTTS
 from sklearn.ensemble import RandomForestClassifier
 from sklearn.preprocessing import LabelEncoder
 
-# Whisper (optional) for transcription
-try:
-    import whisper as openai_whisper
-    whisper_model = openai_whisper.load_model("base", device="cpu")
-except Exception:
-    whisper_model = None
-
-# Translation models (optional)
-try:
-    from transformers import MarianMTModel, MarianTokenizer
-    en_to_ar_model = MarianMTModel.from_pretrained("Helsinki-NLP/opus-mt-en-ar")
-    en_to_ar_tokenizer = MarianTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-ar")
-    ar_to_en_model = MarianMTModel.from_pretrained("Helsinki-NLP/opus-mt-ar-en")
-    ar_to_en_tokenizer = MarianTokenizer.from_pretrained("Helsinki-NLP/opus-mt-ar-en")
-except Exception:
-    en_to_ar_model = en_to_ar_tokenizer = ar_to_en_model = ar_to_en_tokenizer = None
-
-from gtts import gTTS
-
-# ---------------- Configuration ----------------
-CSV_PATH = "deepseek_csv_20251105_09a9e0.csv"  # Use your actual CSV file
+# ======================= Configuration =======================
+CSV_PATH = "deepseek_csv_20251105_09a9e0.csv"
+MULTIMODAL_CSV = "multimodal_audio_dataset.csv"
 AUDIO_FOLDER = "Dataset"
-MODEL_DIR = "models"
-CNN_MODEL_FILE = os.path.join(MODEL_DIR, "ravdess_cnn.h5")
-MODEL_DOWNLOAD_URL = "https://example.com/path/to/ravdess_cnn.h5"  # replace if available
 SAMPLE_RATE = 16000
-N_MFCC = 40
-MAX_MFCC_FRAMES = 128
-EMOTIONS_ALLOWED = ["sad", "angry", "happy", "neutral"]
+EMOTIONS_ALLOWED = ["sad", "happy", "anxious", "neutral", "angry"]
 
-os.makedirs(MODEL_DIR, exist_ok=True)
 os.makedirs(AUDIO_FOLDER, exist_ok=True)
 
-# Diagnostic check
-print("Current working directory:", os.getcwd())
-print("CSV path:", CSV_PATH)
-print("CSV exists:", os.path.exists(CSV_PATH))
-
-# ---------------- Original chatbot lists (kept) ----------------
-MENTAL_KEYWORDS = [
-    "depression", "depressed", "anxiety", "anxious", "panic", "stress", "sad", "lonely",
-    "trauma", "mental", "therapy", "therapist", "counselor", "mood", "overwhelmed", "anger",
-    "fear", "worry", "self-esteem", "confidence", "motivation", "relationship", "cope", "coping",
-    "relax", "calm", "sleep", "emotion", "feeling", "feel", "thoughts", "help", "life", "advice",
-    "unmotivated", "lost", "hopeless", "tired", "burnout", "cry", "hurt", "love", "breakup",
-    "friend", "family", "alone", "heartbroken", "scared", "fearful",
-    "ana", "zahqan", "daye2", "ha2t", "mota3ab", "mota3eb", "za3lan", "malo", "khalni", "mash3or",
-    "bakhaf", "w7ed", "msh 3aref", "mash fahem", "malish", "3ayez", "ayez", "7azeen", "mdaye2",
-    "حزين", "تعبان", "قلق", "خايف", "وحدة", "ضيق", "توتر", "زعلان", "اكتئاب", "علاج",
-    "مشاعر", "مضغوط", "قلقان", "وحدي", "مش مبسوط", "زهقان", "ضايق", "تعب", "مش مرتاح",
-]
+# ======================= Audio Feature Extraction =======================
 
-OFF_TOPIC = [
-    "recipe", "song", "music", "lyrics", "joke", "funny", "laugh", "code", "python", "program",
-    "game", "food", "cook", "movie", "film", "series", "sport", "football", "instagram",
-    "tiktok", "money", "business", "crypto", "ai", "computer",
-    "نكتة", "ضحك", "اغنية", "اغاني", "طبخ", "اكل", "فيلم", "مسلسل", "كورة", "رياضة",
-    "بيزنس", "فلوس", "العاب", "لعبة", "كود", "برمجة", "ذكاء اصطناعي"
-]
+def extract_audio_features(audio_file):
+    """
+    Extract audio features from a .wav file:
+    - Pitch (fundamental frequency)
+    - Energy/Intensity
+    - Tempo (speaking rate)
+    """
+    y, sr = librosa.load(audio_file, sr=SAMPLE_RATE)
+    features = {}
 
-OFF_TOPIC_RESPONSES = [
-    "I'm here to help with emotional and mental well-being. Let's focus on how you're feeling, coping, or managing your emotions today.",
-    "I specialize in mental and emotional health conversations. Tell me what's been on your mind lately.",
-    "Let's bring it back to how you've been feeling — I'm here to help you talk through emotions, stress, or challenges.",
-    "My goal is to support your mental health. How have things been emotionally for you lately?",
-    "I'm here for emotional and mental support only. What's been bothering you recently?",
-]
+    # Pitch features
+    pitches, magnitudes = librosa.piptrack(y=y, sr=sr)
+    pitch_values = [pitches[magnitudes[:, t].argmax(), t] 
+                    for t in range(pitches.shape[1]) 
+                    if magnitudes[:, t].max() > 0]
+
+    features['pitch_mean'] = np.mean(pitch_values) if pitch_values else 0
+    features['pitch_std'] = np.std(pitch_values) if pitch_values else 0
+    features['energy_mean'] = np.mean(librosa.feature.rms(y=y)[0])
+    tempo, _ = librosa.beat.beat_track(y=y, sr=sr)
+    features['tempo'] = tempo
+
+    return features
 
+def generate_audio_training_data(n_samples=100):
+    """
+    Generate synthetic audio dataset for 5 emotions
+    """
+    np.random.seed(42)
+    emotions = ['sad', 'happy', 'anxious', 'neutral', 'angry']
+    data = []
+
+    for emo in emotions:
+        for _ in range(n_samples):
+            if emo == 'sad':
+                pitch_mean = np.random.normal(130, 5)
+                pitch_std = np.random.normal(15, 5)
+                energy_mean = np.random.uniform(0.015, 0.04)
+                tempo = np.random.uniform(70, 90)
+            elif emo == 'happy':
+                pitch_mean = np.random.normal(220, 10)
+                pitch_std = np.random.normal(45, 10)
+                energy_mean = np.random.uniform(0.1, 0.15)
+                tempo = np.random.uniform(100, 130)
+            elif emo == 'anxious':
+                pitch_mean = np.random.normal(180, 10)
+                pitch_std = np.random.normal(60, 10)
+                energy_mean = np.random.uniform(0.06, 0.09)
+                tempo = np.random.uniform(120, 150)
+            elif emo == 'neutral':
+                pitch_mean = np.random.normal(160, 10)
+                pitch_std = np.random.normal(25, 5)
+                energy_mean = np.random.uniform(0.05, 0.08)
+                tempo = np.random.uniform(90, 110)
+            elif emo == 'angry':
+                pitch_mean = np.random.normal(210, 10)
+                pitch_std = np.random.normal(50, 10)
+                energy_mean = np.random.uniform(0.12, 0.18)
+                tempo = np.random.uniform(120, 160)
+            data.append([pitch_mean, pitch_std, energy_mean, tempo, emo])
+
+    df = pd.DataFrame(data, columns=['pitch_mean','pitch_std','energy_mean','tempo','true_emotion'])
+    df.to_csv(MULTIMODAL_CSV, index=False)
+    print("✓ Multimodal audio dataset saved as 'multimodal_audio_dataset.csv'")
+    return df
+
+# ======================= Train Audio RF Model =======================
+if not os.path.exists(MULTIMODAL_CSV):
+    generate_audio_training_data(n_samples=100)
+
+df_audio = pd.read_csv(MULTIMODAL_CSV)
+X_audio = df_audio[['pitch_mean','pitch_std','energy_mean','tempo']]
+y_audio = df_audio['true_emotion']
+
+_audio_rf_le = LabelEncoder()
+y_enc = _audio_rf_le.fit_transform(y_audio)
+
+_audio_rf_model = RandomForestClassifier(n_estimators=200, random_state=42)
+_audio_rf_model.fit(X_audio, y_enc)
+print("✓ Audio RF model trained from multimodal dataset")
+
+# ======================= Emotion Prediction =======================
+def predict_emotion_from_audiofile(audio_filepath):
+    try:
+        features = extract_audio_features(audio_filepath)
+        X = np.array([[features['pitch_mean'], features['pitch_std'], features['energy_mean'], features['tempo']]])
+        pred_enc = _audio_rf_model.predict(X)[0]
+        label = _audio_rf_le.inverse_transform([pred_enc])[0].lower()
+        return label
+    except Exception as e:
+        print(f"Error predicting emotion: {e}")
+        return random.choice(EMOTIONS_ALLOWED)
+
+# ======================= Mental Health Text Chat =======================
+MENTAL_KEYWORDS = ["depression","anxiety","stress","sad","trauma","therapy","mental","emotion","feel","help"]
+OFF_TOPIC = ["song","music","joke","game","food","movie","sport","money","business"]
 MENTAL_RESPONSES_EN = [
-    "I hear that you're going through a difficult time. It takes courage to talk about these feelings.",
-    "Thank you for sharing that with me. Your feelings are completely valid.",
-    "I'm here to listen and support you. Would you like to talk more about what you're experiencing?",
-    "It sounds like you're carrying a lot right now. Remember that you don't have to face everything alone.",
-    "I appreciate you opening up. Let's work through this together, one step at a time.",
-    "Your feelings matter, and I'm here to support you through this challenging time.",
-    "It's okay to not be okay. I'm here to listen whenever you're ready to talk.",
+    "I hear that you're going through a difficult time.",
+    "Thank you for sharing that with me. Your feelings are valid.",
+    "I'm here to listen and support you. Would you like to talk more?",
 ]
 
-MENTAL_RESPONSES_AR = [
-    "أسمع أنك تمر بوقت صعب. يتطلب الأمر شجاعة للتحدث عن هذه المشاعر.",
-    "شكرًا لك على مشاركة ذلك معي. مشاعرك صحيحة تمامًا.",
-    "أنا هنا للاستماع ودعمك. هل ترغب في التحدث أكثر عما تمر به؟",
-    "يبدو أنك تحمل الكثير الآن. تذكر أنك لست مضطرًا لمواجهة كل شيء بمفردك.",
-    "أقدر انفتاحك. دعنا نعمل على هذا معًا، خطوة بخطوة.",
-    "مشاعرك مهمة، وأنا هنا لدعمك خلال هذا الوقت الصعب.",
-    "لا بأس في أن لا تكون على ما يرام. أنا هنا للاستماع عندما تكون مستعدًا للتحدث.",
+OFF_TOPIC_RESPONSES = [
+    "Let's focus on emotional well-being. How are you feeling today?",
+    "I specialize in mental health conversations. Tell me how you're feeling.",
 ]
 
 def contains_arabic(text: str) -> bool:
@@ -118,12 +140,11 @@ def contains_arabic(text: str) -> bool:
 
 def is_mental_health_related(text: str) -> bool:
     text_lower = text.lower()
-    has_arabic = contains_arabic(text_lower)
     if any(word in text_lower for word in OFF_TOPIC):
         return False
     if any(word in text_lower for word in MENTAL_KEYWORDS):
         return True
-    if has_arabic:
+    if contains_arabic(text_lower):
         return True
     return False
 
@@ -133,244 +154,22 @@ def respond(message):
     if not is_mental_health_related(message):
         return random.choice(OFF_TOPIC_RESPONSES)
     if contains_arabic(message):
-        return random.choice(MENTAL_RESPONSES_AR)
+        return "أنا هنا لدعمك، كيف تشعر اليوم؟"
     else:
         return random.choice(MENTAL_RESPONSES_EN)
 
-# ---------------- Audio helpers ----------------
-def load_audio(path, sr=SAMPLE_RATE):
-    if not os.path.isfile(path):
-        raise FileNotFoundError(f"Audio file not found: {path}")
-    data, orig_sr = sf.read(path, dtype='float32')
-    if data.ndim > 1:
-        data = np.mean(data, axis=1)
-    if orig_sr != sr:
-        data = librosa.resample(data, orig_sr, sr)
-    return data
-
-def compute_mfcc_feature(y, sr=SAMPLE_RATE, n_mfcc=N_MFCC, max_len=MAX_MFCC_FRAMES):
-    mfcc = librosa.feature.mfcc(y=y, sr=sr, n_mfcc=n_mfcc)
-    if mfcc.shape[1] < max_len:
-        pad_width = max_len - mfcc.shape[1]
-        mfcc = np.pad(mfcc, ((0,0),(0,pad_width)), mode='constant')
-    else:
-        mfcc = mfcc[:, :max_len]
-    mfcc = (mfcc - np.mean(mfcc)) / (np.std(mfcc) + 1e-9)
-    return mfcc.T  # (time, n_mfcc)
-
-# ---------------- CNN model (RAVDESS-like) ----------------
-def build_cnn_model(input_shape=(MAX_MFCC_FRAMES, N_MFCC), n_classes=4):
-    if not TF_AVAILABLE:
-        raise RuntimeError("TensorFlow not installed. Install tensorflow to use the CNN model.")
-    inp = layers.Input(shape=input_shape)
-    x = layers.Conv1D(64, kernel_size=3, activation='relu', padding='same')(inp)
-    x = layers.BatchNormalization()(x)
-    x = layers.MaxPooling1D(2)(x)
-    x = layers.Conv1D(128, kernel_size=3, activation='relu', padding='same')(x)
-    x = layers.BatchNormalization()(x)
-    x = layers.MaxPooling1D(2)(x)
-    x = layers.Conv1D(256, kernel_size=3, activation='relu', padding='same')(x)
-    x = layers.GlobalAveragePooling1D()(x)
-    x = layers.Dense(128, activation='relu')(x)
-    x = layers.Dropout(0.3)(x)
-    out = layers.Dense(n_classes, activation='softmax')(x)
-    model = models.Model(inputs=inp, outputs=out)
-    model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
-    return model
-
-def download_pretrained_model(url=MODEL_DOWNLOAD_URL, dest=CNN_MODEL_FILE):
-    try:
-        r = requests.get(url, stream=True, timeout=30)
-        r.raise_for_status()
-        with open(dest, 'wb') as f:
-            for chunk in r.iter_content(chunk_size=8192):
-                if chunk:
-                    f.write(chunk)
-        return True
-    except Exception as e:
-        print("Model download failed:", e)
-        return False
-
-# ---------------- RF fallback ----------------
-RF_MODEL_PATH = os.path.join(MODEL_DIR, "rf_emotion.pkl")
-RF_META_PATH = os.path.join(MODEL_DIR, "rf_meta.pkl")
-
-def create_fallback_rf_model():
-    """Create a simple fallback RF model when no dataset is available"""
-    print("Creating fallback RF model with synthetic data...")
-    
-    # Create synthetic MFCC-like features
-    np.random.seed(42)
-    n_samples = 200
-    n_features = N_MFCC
-    
-    X = np.random.randn(n_samples, n_features)
-    emotions = ["sad", "angry", "happy", "neutral"]
-    y = np.random.choice(emotions, n_samples)
-    
-    # Add some pattern to make it somewhat meaningful
-    for i, emotion in enumerate(y):
-        if emotion == "sad":
-            X[i, :5] -= 1.0  # Lower frequencies for sad
-        elif emotion == "angry":
-            X[i, 5:10] += 1.5  # Higher frequencies for angry
-        elif emotion == "happy":
-            X[i, :] += 0.5  # Generally higher for happy
-    
-    le = LabelEncoder()
-    y_enc = le.fit_transform(y)
-    
-    rf = RandomForestClassifier(n_estimators=100, random_state=42)
-    rf.fit(X, y_enc)
-    
-    joblib.dump(rf, RF_MODEL_PATH)
-    joblib.dump({"label_encoder": le}, RF_META_PATH)
-    
-    return rf, {"label_encoder": le}
-
-def train_or_load_rf(csv_path=CSV_PATH, rebuild=False):
-    if os.path.isfile(RF_MODEL_PATH) and not rebuild:
-        try:
-            rf = joblib.load(RF_MODEL_PATH)
-            meta = joblib.load(RF_META_PATH)
-            print("Loaded pre-trained RF model")
-            return rf, meta
-        except Exception as e:
-            print("Error loading saved RF model, rebuilding...", e)
-            rebuild = True
-    
-    if not os.path.isfile(csv_path):
-        print(f"CSV not found at {csv_path}. Creating fallback RF model...")
-        return create_fallback_rf_model()
-    
-    try:
-        df = pd.read_csv(csv_path)
-        if not set(["audio_path", "emotion"]).issubset(df.columns):
-            print("CSV missing required columns, using fallback...")
-            return create_fallback_rf_model()
-        
-        X = []
-        y = []
-        valid_count = 0
-        
-        print("Processing audio files for RF training...")
-        for _, row in df.iterrows():
-            if valid_count >= 100:  # Limit for faster processing
-                break
-                
-            ap = row["audio_path"]
-            if not os.path.isabs(ap):
-                # Try multiple possible locations
-                possible_paths = [
-                    ap,
-                    os.path.join(os.path.dirname(csv_path), ap),
-                    os.path.join(AUDIO_FOLDER, ap),
-                    os.path.join("Dataset", ap)
-                ]
-                ap = None
-                for path in possible_paths:
-                    if os.path.isfile(path):
-                        ap = path
-                        break
-            
-            if not ap or not os.path.isfile(ap):
-                continue
-                
-            try:
-                y_audio = load_audio(ap)
-                feat = compute_mfcc_feature(y_audio).mean(axis=0)  # simple fixed vector
-                X.append(feat)
-                y.append(row["emotion"].lower())
-                valid_count += 1
-                if valid_count % 20 == 0:
-                    print(f"Processed {valid_count} audio files...")
-            except Exception as e:
-                continue
-        
-        if len(X) == 0:
-            print("No valid audio files found, using fallback...")
-            return create_fallback_rf_model()
-        
-        X = np.vstack(X)
-        le = LabelEncoder()
-        y_enc = le.fit_transform(y)
-        
-        rf = RandomForestClassifier(n_estimators=200, random_state=42)
-        rf.fit(X, y_enc)
-        
-        joblib.dump(rf, RF_MODEL_PATH)
-        joblib.dump({"label_encoder": le}, RF_META_PATH)
-        
-        print(f"RF model trained successfully with {len(X)} samples")
-        return rf, {"label_encoder": le}
-        
-    except Exception as e:
-        print(f"Error training RF model: {e}, using fallback...")
-        return create_fallback_rf_model()
-
-# ---------------- On-demand model loader ----------------
-_cnn_model = None
-_rf_model = None
-_rf_meta = None
-_label_map = {i: emo for i, emo in enumerate(EMOTIONS_ALLOWED)}
-
-def prepare_model_on_demand():
-    global _cnn_model, _rf_model, _rf_meta
-    # Try local CNN model if TF available
-    if TF_AVAILABLE and os.path.isfile(CNN_MODEL_FILE):
-        try:
-            _cnn_model = tf.keras.models.load_model(CNN_MODEL_FILE)
-            print("Loaded CNN model")
-            return "cnn"
-        except Exception as e:
-            print("Failed to load local CNN model:", e)
-    # Try to download
-    if TF_AVAILABLE and not os.path.isfile(CNN_MODEL_FILE):
-        try:
-            ok = download_pretrained_model()
-            if ok and os.path.isfile(CNN_MODEL_FILE):
-                _cnn_model = tf.keras.models.load_model(CNN_MODEL_FILE)
-                print("Downloaded and loaded CNN model")
-                return "cnn"
-        except Exception as e:
-            print("Download/load of CNN failed:", e)
-    # Fallback to RF
-    _rf_model, _rf_meta = train_or_load_rf()
-    print("Using RF model for emotion detection")
-    return "rf"
-
-def predict_emotion_from_audiofile(audio_filepath):
-    """
-    Loads model on first use (on-demand), extracts features, and returns one of EMOTIONS_ALLOWED.
-    """
-    global _cnn_model, _rf_model, _rf_meta
-    if _cnn_model is None and _rf_model is None:
-        model_type = prepare_model_on_demand()
-    else:
-        model_type = "cnn" if _cnn_model is not None else "rf"
-
+# ======================= TTS Helper =======================
+def make_tts_for_message(text, lang="en"):
     try:
-        y_audio = load_audio(audio_filepath)
-        
-        if model_type == "cnn" and _cnn_model is not None:
-            mf = compute_mfcc_feature(y_audio)  # (time, n_mfcc)
-            inp = np.expand_dims(mf, axis=0)
-            preds = _cnn_model.predict(inp, verbose=0)
-            idx = int(np.argmax(preds, axis=1)[0])
-            label = _label_map.get(idx, EMOTIONS_ALLOWED[idx % len(EMOTIONS_ALLOWED)])
-            return label
-        else:
-            feat = compute_mfcc_feature(y_audio).mean(axis=0)
-            pred_enc = _rf_model.predict([feat])[0]
-            label = _rf_meta["label_encoder"].inverse_transform([pred_enc])[0]
-            label = label.lower()
-            mapping = {"sadness": "sad", "joy":"happy", "happiness":"happy", "neutral":"neutral", "anger":"angry"}
-            return mapping.get(label, label)
+        tts = gTTS(text, lang=lang)
+        tmp = tempfile.NamedTemporaryFile(suffix=".mp3", delete=False)
+        tts.save(tmp.name)
+        return tmp.name
     except Exception as e:
-        print(f"Error in emotion prediction: {e}")
-        return random.choice(EMOTIONS_ALLOWED)
+        print(f"TTS error: {e}")
+        return None
 
-# ---------------- Supportive short messages (Style 3) ----------------
+# ======================= Combined Voice Chat =======================
 SUPPORT_MESSAGES = {
     "sad": "I'm sorry you're feeling sad. I'm here for you.",
     "angry": "It's okay to feel angry. I'm here to listen.",
@@ -378,100 +177,53 @@ SUPPORT_MESSAGES = {
     "neutral": "Thanks for sharing. I'm here whenever you need to talk."
 }
 
-def make_tts_for_message(text, lang="en"):
-    try:
-        tts = gTTS(text, lang=lang)
-        tmp = tempfile.NamedTemporaryFile(suffix=".mp3", delete=False)
-        tts.save(tmp.name)
-        return tmp.name
-    except Exception as e:
-        print(f"TTS error: {e}")
-        return None
+def load_audio(path, sr=SAMPLE_RATE):
+    if not os.path.isfile(path):
+        raise FileNotFoundError(f"Audio file not found: {path}")
+    data, orig_sr = sf.read(path, dtype='float32')
+    if data.ndim > 1:
+        data = np.mean(data, axis=1)
+    if orig_sr != sr:
+        data = librosa.resample(data, orig_sr, sr)
+    return data
 
-# ---------------- Combined Voice Chat (now with emotion detection) ----------------
 def voice_chat_combined(audio_path, language):
-    """
-    Steps:
-    1) transcribe audio (whisper if available)
-    2) detect emotion from audio tone
-    3) combine meaning + emotion to craft a short supportive message
-    4) return text and TTS audio path
-    """
     if not audio_path:
-        return "No audio received. Please speak into the microphone.", None
+        return "No audio received. Please speak.", None
 
-    # 1) Transcribe
-    user_text = ""
-    try:
-        if whisper_model is not None:
-            res = whisper_model.transcribe(audio_path)
-            user_text = res.get("text", "").strip()
-        else:
-            # fallback: attempt a quick librosa-based silence check and return empty transcription
-            user_text = ""
-    except Exception as e:
-        user_text = ""
-
-    # 2) Emotion detection from tone
-    try:
-        emotion = predict_emotion_from_audiofile(audio_path)
-        print(f"Detected emotion: {emotion}")
-    except Exception as e:
-        print(f"Error detecting emotion: {e}")
-        emotion = random.choice(EMOTIONS_ALLOWED)
-
-    # 3) Craft combined response (short & simple style)
-    emo_cap = emotion.capitalize()
-    support = SUPPORT_MESSAGES.get(emotion, "I hear you. I'm here for you.")
+    # Emotion detection
+    emotion = predict_emotion_from_audiofile(audio_path)
+    print(f"Detected emotion: {emotion}")
     
-    # include a brief echo of user text if available (first 60 chars)
-    if user_text:
-        echo = user_text.strip()
-        if len(echo) > 60:
-            echo = echo[:57].rsplit(" ", 1)[0] + "..."
-        combined_text = f"Detected Emotion: {emo_cap}\n{support}\nYou said: \"{echo}\""
-    else:
-        combined_text = f"Detected Emotion: {emo_cap}\n{support}"
-
-    # 4) TTS (language selection: use Arabic if language == Arabic and gTTS supports it)
-    tts_lang = "ar" if (language and language.lower().startswith("arab")) else "en"
+    support = SUPPORT_MESSAGES.get(emotion, "I hear you. I'm here for you.")
+    tts_lang = "ar" if language.lower().startswith("arab") else "en"
     tts_path = make_tts_for_message(support, lang=tts_lang)
+    
+    return f"Detected Emotion: {emotion.capitalize()}\n{support}", tts_path
 
-    return combined_text, tts_path
-
-# ---------------- Gradio UI ----------------
+# ======================= Gradio UI =======================
 def clear_text():
     return "", ""
 
-with gr.Blocks(title="🧠 Mental Health Therapy Chatbot (Voice + Emotion)") as demo:
+with gr.Blocks(title="🧠 Mental Health Therapy Chatbot") as demo:
     gr.Markdown("# 🧠 Mental Health Therapy Chatbot")
-    gr.Markdown("A supportive space for mental health conversations in English and Arabic")
+    gr.Markdown("Supportive space for mental health conversations (English/Arabic)")
 
     with gr.Tabs():
-        # Text Chat (unchanged)
         with gr.Tab("💬 Text Chat"):
-            gr.Markdown("### Chat with me about how you're feeling")
+            gr.Markdown("### Chat about how you're feeling")
             with gr.Row():
                 with gr.Column():
-                    text_input = gr.Textbox(
-                        label="Type your message here...",
-                        placeholder="How are you feeling today?",
-                        lines=3
-                    )
+                    text_input = gr.Textbox(label="Type your message here...", lines=3)
                     text_submit = gr.Button("Send Message", variant="primary")
                 with gr.Column():
-                    text_output = gr.Textbox(
-                        label="Response",
-                        interactive=False,
-                        lines=5
-                    )
+                    text_output = gr.Textbox(label="Response", interactive=False, lines=5)
             text_submit.click(fn=respond, inputs=[text_input], outputs=[text_output])
             clear_btn = gr.Button("Clear Conversation")
             clear_btn.click(fn=clear_text, outputs=[text_input, text_output])
 
-        # Voice Chat (merged with emotion detection)
         with gr.Tab("🎙️ Voice Chat"):
-            gr.Markdown("### Speak to me in English or Arabic — I'll listen to what you say AND how you say it.")
+            gr.Markdown("### Speak to me — I'll detect emotion and respond")
             with gr.Row():
                 with gr.Column():
                     audio_input_v = gr.Audio(sources=["microphone"], type="filepath", label="🎤 Speak Here")
@@ -480,8 +232,9 @@ with gr.Blocks(title="🧠 Mental Health Therapy Chatbot (Voice + Emotion)") as
                 with gr.Column():
                     voice_output_text = gr.Textbox(label="💬 Chatbot Response (text)", lines=6, interactive=False)
                     voice_output_audio = gr.Audio(label="🔊 Voice Output (TTS)", interactive=False)
-            voice_submit.click(fn=voice_chat_combined, inputs=[audio_input_v, language_input], outputs=[voice_output_text, voice_output_audio])
+            voice_submit.click(fn=voice_chat_combined, inputs=[audio_input_v, language_input],
+                               outputs=[voice_output_text, voice_output_audio])
 
 if __name__ == "__main__":
     print("Starting Mental Health Therapy Chatbot...")
-    demo.launch(share=True)
+    demo.launch(share=True)