Update app.py

app.py

@@ -1,240 +1,138 @@
-# ======================= app.py =======================
-
-import os
-import re
-import random
-import tempfile
-import warnings
-warnings.filterwarnings("ignore")
-
-import numpy as np
-import pandas as pd
-import soundfile as sf
-import librosa
-import joblib
 import gradio as gr
-from gtts import gTTS
-from sklearn.ensemble import RandomForestClassifier
-from sklearn.preprocessing import LabelEncoder
-
-# ======================= Configuration =======================
-CSV_PATH = "deepseek_csv_20251105_09a9e0.csv"
-MULTIMODAL_CSV = "multimodal_audio_dataset.csv"
-AUDIO_FOLDER = "Dataset"
-SAMPLE_RATE = 16000
-EMOTIONS_ALLOWED = ["sad", "happy", "anxious", "neutral", "angry"]
-
-os.makedirs(AUDIO_FOLDER, exist_ok=True)
-
-# ======================= Audio Feature Extraction =======================
-
-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 = {}
-
-    # 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.",
-    "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?",
+from huggingface_hub import InferenceClient
+import random
+import re
+
+# ✅ Allowed mental health keywords (EN + AR + transliterated Arabic)
+MENTAL_KEYWORDS = [
+    # English
+    "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",
+    # Transliterated Arabic
+    "ana", "zahqan", "daye2", "ha2t", "mota3ab", "mota3eb", "za3lan", "malo", "khalni", "mash3or",
+    "bakhaf", "w7ed", "msh 3aref", "mash fahem", "malish", "3ayez", "ayez", "7azeen", "mdaye2",
+    # Arabic
+    "حزين", "تعبان", "قلق", "خايف", "وحدة", "ضيق", "توتر", "زعلان", "اكتئاب", "علاج",
+    "مشاعر", "مضغوط", "قلقان", "وحدي", "مش مبسوط", "زهقان", "ضايق", "تعب", "مش مرتاح",
 ]
 
+# ✅ Off-topic keywords (EN + AR)
+OFF_TOPIC = [
+    # English
+    "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",
+    # Arabic
+    "نكتة", "ضحك", "اغنية", "اغاني", "طبخ", "اكل", "فيلم", "مسلسل", "كورة", "رياضة",
+    "بيزنس", "فلوس", "العاب", "لعبة", "كود", "برمجة", "ذكاء اصطناعي"
+]
+
+# ✅ Random natural off-topic responses
 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.",
+    "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?",
+    "Let's focus on your thoughts and feelings — I can help you process or manage them better.",
+    "It sounds like you might be going off-topic. Can we talk about how you’ve been feeling instead?",
+    "Let’s keep this space focused on your emotions and well-being. What’s been heavy on your mind lately?",
 ]
 
+# ✅ Detect Arabic characters
 def contains_arabic(text: str) -> bool:
     return bool(re.search(r"[\u0600-\u06FF]", text))
 
+# ✅ Function to check if input is related to mental health
 def is_mental_health_related(text: str) -> bool:
     text_lower = text.lower()
+    has_arabic = contains_arabic(text_lower)
+
+    # If message includes off-topic Arabic or English terms → block it
     if any(word in text_lower for word in OFF_TOPIC):
         return False
+
+    # If it has mental-related Arabic/English → allow
     if any(word in text_lower for word in MENTAL_KEYWORDS):
         return True
-    if contains_arabic(text_lower):
+
+    # If purely Arabic but not off-topic → assume emotional (allow)
+    if has_arabic:
         return True
+
+    # Default fallback
     return False
 
-def respond(message):
-    if not message.strip():
-        return "Please type a message first."
+
+# ✅ Main response function
+def respond(
+    message,
+    history: list[dict[str, str]],
+    system_message,
+    max_tokens,
+    temperature,
+    top_p,
+    hf_token: gr.OAuthToken,
+):
     if not is_mental_health_related(message):
-        return random.choice(OFF_TOPIC_RESPONSES)
-    if contains_arabic(message):
-        return "أنا هنا لدعمك، كيف تشعر اليوم؟"
-    else:
-        return random.choice(MENTAL_RESPONSES_EN)
-
-# ======================= TTS Helper =======================
-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
-
-# ======================= 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.",
-    "happy": "I'm glad you're feeling happy. That's good to hear!",
-    "neutral": "Thanks for sharing. I'm here whenever you need to talk."
-}
-
-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 voice_chat_combined(audio_path, language):
-    if not audio_path:
-        return "No audio received. Please speak.", None
-
-    # Emotion detection
-    emotion = predict_emotion_from_audiofile(audio_path)
-    print(f"Detected emotion: {emotion}")
-    
-    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
-
-# ======================= Gradio UI =======================
-def clear_text():
-    return "", ""
-
-with gr.Blocks(title="🧠 Mental Health Therapy Chatbot") as demo:
-    gr.Markdown("# 🧠 Mental Health Therapy Chatbot")
-    gr.Markdown("Supportive space for mental health conversations (English/Arabic)")
-
-    with gr.Tabs():
-        with gr.Tab("💬 Text Chat"):
-            gr.Markdown("### Chat about how you're feeling")
-            with gr.Row():
-                with gr.Column():
-                    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_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])
-
-        with gr.Tab("🎙️ Voice Chat"):
-            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")
-                    language_input = gr.Radio(["English", "Arabic"], value="English", label="Language")
-                    voice_submit = gr.Button("Process Voice", variant="primary")
-                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])
+        yield random.choice(OFF_TOPIC_RESPONSES)
+        return
+
+    locked_system_message = (
+        "You are a licensed mental health therapy assistant. "
+        "You respond with empathy, emotional intelligence, and a therapeutic tone. "
+        "Never answer questions unrelated to emotional or mental wellness, even if they are in another language."
+    )
+
+    client = InferenceClient(token=hf_token.token, model="openai/gpt-oss-20b")
+
+    messages = [{"role": "system", "content": locked_system_message}]
+    messages.extend(history)
+    messages.append({"role": "user", "content": message})
+
+    response = ""
+
+    for message in client.chat_completion(
+        messages,
+        max_tokens=max_tokens,
+        stream=True,
+        temperature=temperature,
+        top_p=top_p,
+    ):
+        choices = message.choices
+        token = ""
+        if len(choices) and choices[0].delta.content:
+            token = choices[0].delta.content
+        response += token
+        yield response
+
+
+# ✅ Gradio interface setup
+chatbot = gr.ChatInterface(
+    respond,
+    type="messages",
+    additional_inputs=[
+        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
+        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
+        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
+        gr.Slider(
+            minimum=0.1,
+            maximum=1.0,
+            value=0.95,
+            step=0.05,
+            label="Top-p (nucleus sampling)",
+        ),
+    ],
+)
+
+with gr.Blocks() as demo:
+    with gr.Sidebar():
+        gr.LoginButton()
+    chatbot.render()
 
 if __name__ == "__main__":
-    print("Starting Mental Health Therapy Chatbot...")
-    demo.launch(share=True)
+    demo.launch()