Update app.py

app.py

@@ -1,24 +1,17 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
-from transformers import pipeline
 import random
 import re
 import torch
-import tempfile
-import soundfile as sf
+from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2Processor
+import librosa
+from gtts import gTTS
 import numpy as np
-import pyttsx3
-
-# ==============================
-# CONFIG
-# ==============================
-
-# Voice emotion model
-emotion_model = pipeline("audio-classification", model="Hatman/audio-emotion-detection")
-
-# Text model setup
-MENTAL_MODEL = "tanusrich/Mental_Health_Chatbot"
+import tempfile
+import os
 
+# ------------------------------
+# Allowed mental health keywords
 MENTAL_KEYWORDS = [
     "depression", "depressed", "anxiety", "anxious", "panic", "stress", "sad", "lonely",
     "trauma", "mental", "therapy", "therapist", "counselor", "mood", "overwhelmed", "anger",
@@ -26,12 +19,13 @@ MENTAL_KEYWORDS = [
     "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", "za3lan", "malo", "khalni", "mash3or", "bakhaf",
-    "w7ed", "msh 3aref", "mash fahem", "malish", "3ayez", "ayez", "7azeen", "mdaye2",
+    "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",
 ]
 
+# Off-topic keywords
 OFF_TOPIC = [
     "recipe", "song", "music", "lyrics", "joke", "funny", "laugh", "code", "python", "program",
     "game", "food", "cook", "movie", "film", "series", "sport", "football", "instagram",
@@ -40,17 +34,17 @@ OFF_TOPIC = [
     "بيزنس", "فلوس", "العاب", "لعبة", "كود", "برمجة", "ذكاء اصطناعي"
 ]
 
+# Random off-topic responses
 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 to help with emotional and mental well-being. Let's focus on how you're feeling today.",
+    "I specialize in mental and emotional health conversations. Tell me what’s on your mind.",
+    "Let’s bring it back to your feelings — I’m here to help you talk through emotions or stress.",
+    "My goal is to support your mental health. How have things been emotionally lately?",
+    "I’m here for emotional support only. What’s been bothering you recently?",
 ]
 
-# ==============================
-# HELPERS
-# ==============================
-
+# ------------------------------
+# Arabic detection
 def contains_arabic(text: str) -> bool:
     return bool(re.search(r"[\u0600-\u06FF]", text))
 
@@ -64,83 +58,69 @@ def is_mental_health_related(text: str) -> bool:
         return True
     return False
 
-# Text-to-speech helper
-def speak_text(text):
-    engine = pyttsx3.init()
-    engine.setProperty("rate", 175)
-    engine.say(text)
-    engine.runAndWait()
-
-# ==============================
-# MAIN CHAT RESPONSE
-# ==============================
-
-def respond(
-    message,
-    history,
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-    hf_token: gr.OAuthToken,
-):
+# ------------------------------
+# Load voice emotion model
+voice_model_name = "Hatman/audio-emotion-detection"
+voice_model = Wav2Vec2ForSequenceClassification.from_pretrained(voice_model_name)
+voice_processor = Wav2Vec2Processor.from_pretrained(voice_model_name)
+
+def detect_voice_emotion(audio_file):
+    audio, sr = librosa.load(audio_file, sr=16000)
+    inputs = voice_processor(audio, sampling_rate=16000, return_tensors="pt", padding=True)
+    with torch.no_grad():
+        logits = voice_model(**inputs).logits
+    predicted_id = torch.argmax(logits, dim=-1).item()
+    return voice_model.config.id2label[predicted_id]
+
+# ------------------------------
+# Chat + voice response with TTS
+def respond(message, history: list[dict[str, str]], system_message, max_tokens, temperature, top_p, hf_token: gr.OAuthToken, audio=None):
+    transcript = {"user": "", "bot": "", "tts": None}
+    response_text = ""
+    
+    # Detect audio emotion
+    if audio:
+        emotion = detect_voice_emotion(audio)
+        response_text += f"[Detected mood: {emotion}] "
+    
+    transcript["user"] = message
     if not is_mental_health_related(message):
-        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."
-    )
-
-    client = InferenceClient(token=hf_token.token, model=MENTAL_MODEL)
-
-    messages = [{"role": "system", "content": locked_system_message}]
-    messages.extend(history)
-    messages.append({"role": "user", "content": message})
-
-    response = ""
-
-    for msg in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = msg.choices[0].delta.content if len(msg.choices) and msg.choices[0].delta else ""
-        response += token
-        yield response
-
-    # Auto speak last reply
-    speak_text(response)
-
-
-# ==============================
-# VOICE EMOTION DETECTION
-# ==============================
-
-def detect_mood(audio_file):
-    if audio_file is None:
-        return "No audio detected."
-    result = emotion_model(audio_file)
-    mood = max(result, key=lambda x: x["score"])["label"]
-    return f"Detected mood: {mood}"
-
-
-# ==============================
-# UI
-# ==============================
-
-with gr.Blocks(theme=gr.themes.Soft(), title="Mental Health Therapy Chatbot") as app:
-    gr.Markdown(
-        "<h1 style='text-align:center; color:#3B82F6;'>🧠 Mental Health Therapy Chatbot</h1>"
-        "<p style='text-align:center;'>Text or talk about how you feel — your emotional support companion.</p>"
-    )
-
+        response_text += random.choice(OFF_TOPIC_RESPONSES)
+        transcript["bot"] = response_text
+    else:
+        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})
+        
+        # Stream response
+        for msg in client.chat_completion(messages, max_tokens=max_tokens, stream=True, temperature=temperature, top_p=top_p):
+            choices = msg.choices
+            token = ""
+            if len(choices) and choices[0].delta.content:
+                token = choices[0].delta.content
+            response_text += token
+        transcript["bot"] = response_text
+    
+    # Convert response to speech (TTS)
+    tts = gTTS(text=transcript["bot"], lang='ar' if contains_arabic(transcript["bot"]) else 'en')
+    tmp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3")
+    tts.save(tmp_file.name)
+    transcript["tts"] = tmp_file.name
+    
+    return transcript
+
+# ------------------------------
+# Gradio UI
+with gr.Blocks() as demo:
     with gr.Row():
-        with gr.Column(scale=2):
+        with gr.Column():
             chatbot = gr.ChatInterface(
                 respond,
                 type="messages",
@@ -148,23 +128,9 @@ with gr.Blocks(theme=gr.themes.Soft(), title="Mental Health Therapy Chatbot") as
                     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)"
-                    ),
+                    gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
+                    gr.Audio(source="microphone", type="filepath", label="Record your voice (optional)"),
+                    gr.OAuthToken(label="Hugging Face Token"),
                 ],
             )
-        with gr.Column(scale=1):
-            gr.Markdown("### 🎤 Voice Mood Detection")
-            audio_input = gr.Audio(sources=["microphone"], type="filepath", label="Record or Upload Voice")
-            mood_output = gr.Textbox(label="Detected Mood", interactive=False)
-            audio_btn = gr.Button("Analyze Mood")
-            audio_btn.click(detect_mood, inputs=audio_input, outputs=mood_output)
-
-    with gr.Row():
-        gr.Markdown("<hr>")
-        gr.Markdown(
-            "<p style='text-align:center;'>Made with ❤️ to support emotional wellness. — Panto</p>"
-        )
-
-if __name__ == "__main__":
-    app.launch()
+    demo.launch()