app.py

app.py

@@ -1,3 +1,4 @@
+# app.py
 import gradio as gr
 import wikipedia
 import numpy as np
@@ -6,13 +7,12 @@ from langdetect import detect
 from gtts import gTTS
 from transformers import pipeline
 from sentence_transformers import SentenceTransformer
-import tempfile, os
-import torch
+import tempfile
 import speech_recognition as sr
-from functools import lru_cache
 from pydub import AudioSegment
+from functools import lru_cache
 
-# ===== Model Setup =====
+# --- Load models ---
 models = {}
 def load_models():
     models['encoder'] = SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2')
@@ -20,10 +20,9 @@ def load_models():
     for lang in ['fr', 'ar', 'zh', 'es']:
         models[f'en_to_{lang}'] = pipeline('translation_en_to_' + lang, model=f'Helsinki-NLP/opus-mt-en-{lang}')
     models['answer_gen'] = pipeline('text2text-generation', model='google/flan-t5-base', max_length=1024)
-
 load_models()
 
-# ===== Utility Functions =====
+# --- Utility functions ---
 def detect_language(text):
     try:
         return detect(text)
@@ -41,108 +40,121 @@ def translate(text, src, tgt):
 
 def tts_play(text, lang):
     tts = gTTS(text=text, lang=lang)
-    path = tempfile.mktemp(suffix=".mp3")
+    path = tempfile.mktemp(suffix='.mp3')
     tts.save(path)
     return path
 
 def chunk_text(text, max_words=100):
     sentences = text.split('. ')
-    chunks, current_chunk, current_len = [], [], 0
+    chunks, current, length = [], [], 0
     for sent in sentences:
         words = sent.split()
-        if current_len + len(words) > max_words:
-            chunks.append('. '.join(current_chunk))
-            current_chunk = [sent]
-            current_len = len(words)
+        if length + len(words) > max_words:
+            chunks.append(' '.join(current))
+            current, length = [sent], len(words)
         else:
-            current_chunk.append(sent)
-            current_len += len(words)
-    if current_chunk:
-        chunks.append('. '.join(current_chunk))
+            current.append(sent)
+            length += len(words)
+    if current: chunks.append(' '.join(current))
     return chunks
 
 def build_faiss_index(chunks, model):
-    embeddings = model.encode(chunks, convert_to_numpy=True)
-    index = faiss.IndexFlatL2(embeddings.shape[1])
-    index.add(embeddings)
+    emb = model.encode(chunks, convert_to_numpy=True)
+    index = faiss.IndexFlatL2(emb.shape[1])
+    index.add(emb)
     return index
 
 @lru_cache(maxsize=20)
 def prepare_faiss_for_topic(topic):
     wikipedia.set_lang('en')
     page = wikipedia.page(topic)
-    content = page.content[:5000]
-    chunks = chunk_text(content)
-    index = build_faiss_index(chunks, models['encoder'])
-    return chunks, index
-
-def retrieve_context(question, index, chunks, model, top_k=5):
-    q_emb = model.encode([question], convert_to_numpy=True)
-    _, indices = index.search(q_emb, top_k)
-    return ' '.join([chunks[i] for i in indices[0]])
-
-# ===== Main Inference Function =====
-def qa_system(audio, text_question, topic, output_lang):
-    question = ""
-    if audio is not None:
+    chunks = chunk_text(page.content)  # Use full content, no slicing limit
+    return chunks, build_faiss_index(chunks, models['encoder'])
+
+def retrieve_context(q, idx, chunks, model, top_k=5):
+    emb = model.encode([q], convert_to_numpy=True)
+    _, inds = idx.search(emb, top_k)
+    return ' '.join(chunks[i] for i in inds[0])
+
+# --- Main Q&A function ---
+def qa_system(audio, text_q, topic, lang, history):
+    q = ''
+    if audio:
         try:
             r = sr.Recognizer()
-            audio_wav_path = tempfile.mktemp(suffix=".wav")
-            sound = AudioSegment.from_file(audio)
-            sound.export(audio_wav_path, format="wav")
-            with sr.AudioFile(audio_wav_path) as source:
-                audio_data = r.record(source)
-                question = r.recognize_google(audio_data)
+            wav = tempfile.mktemp('.wav')
+            AudioSegment.from_file(audio).export(wav, format='wav')
+            with sr.AudioFile(wav) as src:
+                q = r.recognize_google(r.record(src))
         except Exception as e:
-            return f"❌ Could not understand the audio: {e}", None, None
-    elif text_question:
-        question = text_question.strip()
+            return f"❌ Could not transcribe audio: {e}", None, history, ''
+    elif text_q:
+        q = text_q.strip()
     else:
-        return "❌ Please provide a voice or text question.", None, None
-
-    input_lang = detect_language(question)
+        return '❌ Please speak or type your question.', None, history, ''
 
     try:
-        chunks, faiss_index = prepare_faiss_for_topic(topic)
-    except:
-        return "❌ Error loading topic from Wikipedia.", None, None
-
-    context = retrieve_context(question, faiss_index, chunks, models['encoder'], top_k=5)
-    question_en = translate(question, input_lang, 'en')
-    prompt = f"Answer based on the context:\nContext: {context}\nQuestion: {question_en}"
-    answer_en = models['answer_gen'](prompt)[0]['generated_text']
-
-    if output_lang == 'en':
-        answer = answer_en
-    elif output_lang == 'am':
-        answer = "Amharic translation not supported."
-    else:
-        answer = translate(answer_en, 'en', output_lang)
-
-    audio_path = tts_play(answer, output_lang)
-    return f"You asked: {question}\n\nAnswer: {answer}", audio_path, answer
-
-# ===== Gradio UI =====
-lang_options = ['en', 'am', 'fr', 'ar', 'es', 'zh']
-
-demo = gr.Interface(
-    fn=qa_system,
-    inputs=[
-        gr.Audio(type="filepath", label="🎤 Ask your Question by Voice (optional)"),
-        gr.Textbox(label="✍️ Or type your Question here (optional)"),
-        gr.Textbox(value="Artificial intelligence", label="📚 Wikipedia Topic"),
-        gr.Dropdown(choices=lang_options, value='en', label="🌍 Output Language")
-    ],
-    outputs=[
-        gr.Textbox(label="🤖 Answer Output"),
-        gr.Audio(label="🔊 Answer Playback"),
-        gr.Textbox(label="📝 Translated Answer Text")
-    ],
-    title="🌍 Multilingual Voice/Text Q&A Assistant",
-    description="""
-    <h3 style='text-align: center; font-weight: bold; font-style: italic;'>👋 Welcome to the Multilingual Wikipedia Q&A Assistant</h3>
-    <p style='text-align: center;'>Ask questions by voice or text in different languages, and get spoken and translated answers using AI + Wikipedia.</p>
-    """
-)
-
-demo.launch()
+        chunks, idx = prepare_faiss_for_topic(topic)
+    except Exception as e:
+        return f'Error loading content: {e}', None, history, ''
+
+    ctx = retrieve_context(q, idx, chunks, models['encoder'])
+    q_en = translate(q, detect_language(q), 'en')
+
+    # Debug prints — remove in production
+    print("Question (original):", q)
+    print("Question (English):", q_en)
+    print("Retrieved context snippet:", ctx[:500], "...\n")
+
+    prompt = f"Context:\n{ctx}\n\nQuestion: {q_en}\nAnswer:"
+    ans_en = models['answer_gen'](prompt)[0]['generated_text']
+
+    print("Generated answer (English):", ans_en)
+
+    ans = ans_en if lang == 'en' else translate(ans_en, 'en', lang)
+    audio_path = tts_play(ans, lang)
+    history.append((q, ans))
+    chat = '\n\n'.join(f"Q{i+1}: {x}\nA{i+1}: {y}" for i,(x,y) in enumerate(history))
+    return f'You asked: {q}\n\nAnswer: {ans}', audio_path, history, chat
+
+def clear_all():
+    return None, '', None, [], ''
+
+# --- Gradio UI with styling ---
+css_style = """
+.gradio-container {
+    background-color: #cce7ff !important;  /* Light blue */
+    border: 3px solid #000022 !important; /* Balanced blue-black border */
+    border-radius: 12px;
+    padding: 20px;
+}
+"""
+
+with gr.Blocks(css=css_style) as demo:
+    gr.Markdown("""
+    <h1 style='color:#003366; text-align:center; margin-bottom: 0;'>🌐 Multilingual Wikipedia Q&A Assistant</h1>
+    <p style='text-align:center; font-size:16px; margin-top: 0;'>Ask your questions by typing or speaking, and get answers in your language!</p>
+    """)
+
+    state = gr.State([])
+
+    with gr.Row():
+        ai = gr.Audio(type='filepath', label='🎤 Speak your question')
+        ti = gr.Textbox(lines=3, placeholder='Or type your question here')
+
+    with gr.Row():
+        tp = gr.Textbox(value='Artificial intelligence', label='Wikipedia Topic')
+        lg = gr.Dropdown(['en','am','fr','ar','es','zh'], value='en', label='Output Language')
+
+    with gr.Row():
+        sb = gr.Button('🔍 Get Answer')
+        cb = gr.Button('🗑️ Clear All')
+
+    ao = gr.Textbox(label='🤖 Answer')
+    av = gr.Audio(label='🔊 Listen Answer')
+    cd = gr.Markdown(label='🗂️ Chat History')
+
+    sb.click(qa_system, inputs=[ai, ti, tp, lg, state], outputs=[ao, av, state, cd])
+    cb.click(clear_all, outputs=[ai, ti, tp, state, cd])
+
+demo.launch(share=True)