Update app.py

app.py

@@ -3,21 +3,21 @@ from huggingface_hub import InferenceClient
 import random
 import re
 import torch
-from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2Processor
+from transformers import Wav2Vec2ForSequenceClassification, Wav2Vec2FeatureExtractor
 import librosa
 from gtts import gTTS
 import tempfile
 import os
 
-# ===== Mental health keywords (EN + AR + transliterated AR)
+# ========== MENTAL HEALTH FILTERS ==========
+
 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",
+    "trauma", "mental", "therapy", "therapist", "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",
@@ -31,7 +31,6 @@ 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",
-    # Arabic
     "نكتة", "ضحك", "اغنية", "اغاني", "طبخ", "اكل", "فيلم", "مسلسل", "كورة", "رياضة",
     "بيزنس", "فلوس", "العاب", "لعبة", "كود", "برمجة", "ذكاء اصطناعي"
 ]
@@ -42,10 +41,11 @@ OFF_TOPIC_RESPONSES = [
     "Let's bring it back to your emotions — I'm here to help process stress or challenges.",
 ]
 
-# Detect Arabic text
+
 def contains_arabic(text: str) -> bool:
     return bool(re.search(r"[\u0600-\u06FF]", text))
 
+
 def is_mental_health_related(text: str) -> bool:
     text_lower = text.lower()
     if any(word in text_lower for word in OFF_TOPIC):
@@ -56,84 +56,82 @@ def is_mental_health_related(text: str) -> bool:
         return True
     return False
 
-# ===== Voice emotion detection
+
+# ========== EMOTION DETECTION MODEL ==========
+
 voice_model_name = "Hatman/audio-emotion-detection"
 voice_model = Wav2Vec2ForSequenceClassification.from_pretrained(voice_model_name)
-voice_processor = Wav2Vec2Processor.from_pretrained(voice_model_name)
+feature_extractor = Wav2Vec2FeatureExtractor.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)
+    inputs = feature_extractor(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 function with mood, TTS, transcript
+
+# ========== RESPONSE LOGIC ==========
+
 def respond(message, history, system_message, max_tokens, temperature, top_p, hf_token: gr.OAuthToken, audio=None):
     transcript = []
     response_text = ""
 
+    # Mood detection from voice
     if audio:
         mood = detect_voice_emotion(audio)
         response_text += f"[Detected mood: {mood}] "
 
+    # Mental health filtering
     if not is_mental_health_related(message):
-        response_text += random.choice(OFF_TOPIC_RESPONSES)
-        transcript.append(("User", message))
-        transcript.append(("Bot", response_text))
+        bot_reply = random.choice(OFF_TOPIC_RESPONSES)
+        transcript.extend([("User", message), ("Bot", bot_reply)])
         tts_file = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3").name
-        tts = gTTS(response_text)
-        tts.save(tts_file)
-        return response_text, tts_file, transcript
+        gTTS(bot_reply).save(tts_file)
+        return bot_reply, tts_file, transcript
 
+    # GPT-based mental health conversation
+    client = InferenceClient(token=hf_token.token, model="openai/gpt-oss-20b")
     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."
+        "Never answer unrelated questions."
     )
 
-    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})
 
     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
+        if msg.choices and msg.choices[0].delta.content:
+            response_text += msg.choices[0].delta.content
 
-    transcript.append(("User", message))
-    transcript.append(("Bot", response_text))
+    transcript.extend([("User", message), ("Bot", response_text)])
 
     tts_file = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3").name
-    tts = gTTS(response_text)
-    tts.save(tts_file)
+    gTTS(response_text).save(tts_file)
 
     return response_text, tts_file, transcript
 
-# ===== Gradio UI
+
+# ========== GRADIO UI ==========
+
 with gr.Blocks() as demo:
-    gr.Markdown("## Mental Health Chatbot with Voice Mood Detection")
-    with gr.Row():
-        with gr.Column():
-            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)"),
-                    gr.Audio(label="Record your voice (optional)", type="filepath"),
-                    gr.OAuthToken(label="Hugging Face Token"),
-                ],
-            )
-
-    # Output area for transcript
-    transcript_box = gr.Textbox(label="Transcript (User & Bot)", interactive=False)
-
-demo.launch()
+    gr.Markdown("## 🧠 Mental Health Chatbot with Voice Mood Detection & TTS")
+    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"),
+            gr.Audio(label="🎙️ Speak (optional)", type="filepath"),
+            gr.OAuthToken(label="Hugging Face Token"),
+        ],
+    )
+
+demo.launch()