Update app.py

app.py

@@ -1,923 +1,923 @@
-# Specific conda environment : Kabyle_streamlit
-
-# Streamlit TTS Reader - Taqbaylit TTS Sɣer Adlis
-# Bouaziz Ait Driss October 2025
-
-import streamlit as st
-import fitz  # PyMuPDF
-import re
-import numpy as np
-import torch
-from scipy.io.wavfile import write as wav_write
-import tempfile
-import os
-import base64
-import json
-import io
-import threading
-import queue
-import time
-import pdfplumber
-
-# Configure Streamlit page
-st.set_page_config(
-    page_title="Kabyle TTS Document Reader",
-    page_icon="🎙️",
-    layout="wide",
-    initial_sidebar_state="expanded",
-)
-
-# Try to import transformers
-try:
-    from transformers import VitsModel, AutoTokenizer
-    TRANSFORMERS_AVAILABLE = True
-except ImportError:
-    try:
-        from transformers import AutoModel, AutoTokenizer
-        TRANSFORMERS_AVAILABLE = True
-        VitsModel = AutoModel
-    except ImportError:
-        TRANSFORMERS_AVAILABLE = False
-
-# --- Model Loading ---
-@st.cache_resource
-def load_model():
-    if not TRANSFORMERS_AVAILABLE:
-        return None, None, "cpu"
-    try:
-        model = VitsModel.from_pretrained("facebook/mms-tts-kab")
-        tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-kab")
-        device = "cuda" if torch.cuda.is_available() else "cpu"
-        model = model.to(device)
-        return model, tokenizer, device
-    except Exception as e:
-        st.error(f"Error loading model: {e}")
-        return None, None, "cpu"
-
-model, tokenizer, device = load_model()
-
-# --- Text Processing ---
-def clean_text(text):
-    if not text:
-        return text
-    text = re.sub(r' +', ' ', text)
-    text = re.sub(r'\n\s*\n', '\n\n', text)
-    return text.strip()
-
-def smart_split_paragraphs(text, initial_paragraphs=5, initial_word_target=50, normal_word_target=200):
-    """
-    Smart paragraph splitting strategy:
-    - First 'initial_paragraphs' paragraphs: ~initial_word_target words each (for quick startup)
-    - Remaining paragraphs: ~normal_word_target words each (for better timing)
-    """
-    if not text:
-        return []
-    
-    # First, split by major paragraph breaks
-    raw_paragraphs = re.split(r'\n\s*\n', text)
-    raw_paragraphs = [p.strip() for p in raw_paragraphs if p.strip()]
-    
-    if not raw_paragraphs:
-        return []
-    
-    final_paragraphs = []
-    current_sentences = []
-    
-    # Split each raw paragraph into sentences
-    all_sentences = []
-    for paragraph in raw_paragraphs:
-        # Split on sentence endings: . ! ? : ; followed by space
-        sentences = re.split(r'(?<=[.!?:;])\s+', paragraph)
-        sentences = [s.strip() for s in sentences if s.strip()]
-        all_sentences.extend(sentences)
-    
-    if not all_sentences:
-        return []
-    
-    # Build initial quick-start paragraphs (shorter)
-    word_count = 0
-    for sentence in all_sentences:
-        sentence_word_count = len(sentence.split())
-        
-        # For first few paragraphs, use smaller target
-        if len(final_paragraphs) < initial_paragraphs:
-            target_word_count = initial_word_target
-        else:
-            target_word_count = normal_word_target
-        
-        # If adding this sentence would exceed target and we have some content, start new paragraph
-        if current_sentences and word_count + sentence_word_count > target_word_count:
-            # Join current sentences to form a paragraph
-            paragraph_text = ' '.join(current_sentences)
-            final_paragraphs.append(paragraph_text)
-            current_sentences = [sentence]
-            word_count = sentence_word_count
-        else:
-            current_sentences.append(sentence)
-            word_count += sentence_word_count
-    
-    # Add the last paragraph if we have remaining sentences
-    if current_sentences:
-        paragraph_text = ' '.join(current_sentences)
-        final_paragraphs.append(paragraph_text)
-    
-    return final_paragraphs
-
-def split_paragraph_into_phrases(paragraph, max_phrase_length=150):
-    if not paragraph:
-        return []
-    
-    phrases = []
-    current_phrase = ""
-    sentences = re.split(r'(?<=[.!?])\s+', paragraph)
-    
-    for sentence in sentences:
-        if len(current_phrase) + len(sentence) > max_phrase_length and current_phrase:
-            phrases.append(current_phrase.strip())
-            current_phrase = sentence
-        else:
-            current_phrase = current_phrase + " " + sentence if current_phrase else sentence
-
-    if current_phrase.strip():
-        phrases.append(current_phrase.strip())
-
-    return phrases
-
-def generate_audio_with_precise_timing(paragraph):
-    """Generate audio with precise phrase-level timing using per-phrase generation"""
-    
-    if model is None or tokenizer is None:
-        return create_dummy_audio_for_paragraph(paragraph), 16000, []
-
-    try:
-        phrases = split_paragraph_into_phrases(paragraph)
-        if not phrases:
-            return create_dummy_audio_for_paragraph(paragraph), 16000, []
-
-        all_audio = []
-        phrase_durations = []
-        sampling_rate = 16000
-        
-        # Generate audio for each phrase separately to get exact timing
-        for i, phrase in enumerate(phrases):
-            if not phrase.strip():
-                continue
-                
-            # Generate audio for this specific phrase
-            inputs = tokenizer(phrase, return_tensors="pt").to(device)
-            with torch.no_grad():
-                output = model(**inputs)
-            
-            # Extract audio data
-            if hasattr(output, 'waveform'):
-                audio_chunk = output.waveform.cpu().numpy().squeeze()
-            else:
-                audio_chunk = output[0].cpu().numpy().squeeze() if len(output) > 0 else None
-            
-            if audio_chunk is not None:
-                # Calculate exact duration for this phrase
-                phrase_duration = len(audio_chunk) / sampling_rate
-                phrase_durations.append(phrase_duration)
-                
-                # Normalize and store
-                max_val = np.max(np.abs(audio_chunk))
-                if max_val > 0:
-                    audio_chunk = audio_chunk / max_val * 0.8
-                all_audio.append(audio_chunk)
-            else:
-                # Fallback: estimate duration if audio generation failed
-                word_count = len(phrase.split())
-                estimated_duration = word_count * 0.4
-                phrase_durations.append(estimated_duration)
-
-        if all_audio:
-            # Concatenate all phrase audio
-            full_audio = np.concatenate(all_audio)
-            total_duration = len(full_audio) / sampling_rate
-            
-            # Build precise timing info using actual phrase durations
-            timing_info = []
-            current_time = 0.0
-            
-            for i, (phrase, duration) in enumerate(zip(phrases, phrase_durations)):
-                timing_info.append({
-                    'text': phrase,
-                    'start': current_time,
-                    'end': current_time + duration,
-                    'duration': duration
-                })
-                current_time += duration
-            
-            return full_audio, sampling_rate, timing_info
-        else:
-            return create_dummy_audio_for_paragraph(paragraph), 16000, []
-            
-    except Exception as e:
-        return create_dummy_audio_for_paragraph(paragraph), 16000, []
-
-def create_dummy_audio_for_paragraph(paragraph):
-    word_count = len(paragraph.split())
-    total_duration = word_count * 0.4
-    sampling_rate = 16000
-    t = np.linspace(0, total_duration, int(sampling_rate * total_duration))
-    audio = 0.3 * np.sin(2 * np.pi * 220 * t)
-    return audio
-
-# Add a fallback to plumber for reading diacritic letters
-def read_document(file_path):
-    text = ""
-    try:
-        if file_path.lower().endswith('.pdf'):
-            # FIRST TRY: PyMuPDF (faster for digital PDFs)
-            try:
-                pdf_document = fitz.open(file_path)
-                for page_num in range(len(pdf_document)):
-                    page = pdf_document[page_num]
-                    text += page.get_text() + "\n"
-                pdf_document.close()
-                
-                # Check if we got meaningful text with diacritics
-                if text.strip() and any(char in text for char in ['é', 'è', 'à', 'ù', 'ï', 'ö', 'α', 'β', 'γ']):
-                    return text, []
-                    
-            except Exception as e:
-                text = ""  # Reset text if PyMuPDF fails
-                
-            # FALLBACK: pdfplumber (better for scanned/diacritic PDFs)
-            try:
-                import pdfplumber
-                with pdfplumber.open(file_path) as pdf:
-                    text = ""
-                    for page in pdf.pages:
-                        page_text = page.extract_text() or ""
-                        text += page_text + "\n"
-            except ImportError:
-                return "pdfplumber not available", []
-            except Exception as e:
-                return f"Both PDF methods failed: {e}", []
-                
-        elif file_path.lower().endswith('.txt'):
-            with open(file_path, 'r', encoding='utf-8') as f:
-                text = f.read()
-        else:
-            return "Unsupported file format", []
-            
-    except Exception as e:
-        return f"Error reading document: {e}", []
-        
-    return text, []
-
-def normalize_audio(audio_data):
-    if audio_data.dtype != np.float32:
-        audio_data = audio_data.astype(np.float32)
-    max_val = np.max(np.abs(audio_data))
-    if max_val > 0:
-        audio_data = audio_data / max_val * 0.9
-    return audio_data
-
-def get_audio_bytes(audio_data):
-    audio_bytes = io.BytesIO()
-    audio_data = normalize_audio(audio_data)
-    audio_int16 = np.clip(audio_data * 32767, -32768, 32767).astype(np.int16)
-    wav_write(audio_bytes, 16000, audio_int16)
-    audio_bytes.seek(0)
-    return audio_bytes.getvalue()
-
-# --- Background Audio Generation ---
-def background_audio_worker(paragraphs, audio_queue, start_index=0):
-    """Background worker to generate audio for paragraphs - NO SESSION STATE ACCESS"""
-    try:
-        for i, paragraph in enumerate(paragraphs):
-            paragraph_index = start_index + i
-            word_count = len(paragraph.split())
-            
-            # Generate audio with precise timing
-            audio_data, sampling_rate, timing_info = generate_audio_with_precise_timing(paragraph)
-            
-            # Send to main thread via queue ONLY
-            audio_queue.put({
-                'paragraph_index': paragraph_index,
-                'paragraph_text': paragraph,
-                'audio_data': audio_data,
-                'timing_info': timing_info,
-                'audio_duration': len(audio_data) / sampling_rate,
-                'audio_bytes': get_audio_bytes(audio_data),
-                'word_count': word_count
-            })
-            
-    except Exception as e:
-        pass
-
-# --- Queue Processing ---
-def process_audio_queue(audio_queue):
-    """Process all available items in the audio queue - called from main thread only"""
-    processed_count = 0
-    try:
-        while True:
-            # Non-blocking check
-            audio_data = audio_queue.get_nowait()
-            
-            # SAFE: Main thread updating session state
-            st.session_state.paragraphs_data[audio_data['paragraph_index']] = audio_data
-            processed_count += 1
-            
-    except queue.Empty:
-        pass
-    
-    return processed_count
-
-# Initialize session state
-if 'processed' not in st.session_state:
-    st.session_state.processed = False
-if 'current_paragraph_index' not in st.session_state:
-    st.session_state.current_paragraph_index = 0
-if 'total_paragraphs' not in st.session_state:
-    st.session_state.total_paragraphs = 0
-if 'paragraphs_data' not in st.session_state:
-    st.session_state.paragraphs_data = {}
-if 'audio_ready' not in st.session_state:
-    st.session_state.audio_ready = False
-if 'background_worker_started' not in st.session_state:
-    st.session_state.background_worker_started = False
-if 'reading_status' not in st.session_state:
-    st.session_state.reading_status = "Ready to start reading"
-if 'current_document_id' not in st.session_state:
-    st.session_state.current_document_id = None
-if 'audio_queue' not in st.session_state:
-    st.session_state.audio_queue = queue.Queue()
-if 'playback_speed' not in st.session_state:
-    st.session_state.playback_speed = 1.0  # Default normal speed
-if 'full_document_text' not in st.session_state:
-    st.session_state.full_document_text = ""
-if 'smart_splitting_expanded' not in st.session_state:
-    st.session_state.smart_splitting_expanded = False
-
-# --- Streamlit UI ---
-def main():
-    # Custom CSS for better styling - FIXED FONT HARMONIZATION
-    st.markdown("""
-    <style>
-    .main-title {
-        font-size: 1.5rem !important;
-        font-weight: bold !important;
-        margin-bottom: -1rem !important;
-        margin-top: -1rem !important;      /* MAXIMUM REDUCED top margin */
-        color: #1f77b4;
-    }
-    .section-title {
-        font-size: 1.3rem !important;
-        font-weight: bold !important;
-        margin-bottom: 0.3rem !important;
-        margin-top: 0rem !important;
-        color: #2e86ab;
-    }
-    .document-viewer {
-        max-height: 70vh;
-        overflow-y: auto;
-        border: 1px solid #e1e1e1;
-        border-radius: 8px;
-        padding: 15px;
-        background: #fafafa;
-        font-family: Arial, sans-serif;
-        line-height: 1.6;
-        scrollbar-width: thin;
-        scrollbar-color: #888 #f1f1f1;
-        color: #333333;  /* ADD THIS LINE - ensures dark text */
-    }
-    .document-viewer::-webkit-scrollbar {
-        width: 8px;
-    }
-    .document-viewer::-webkit-scrollbar-track {
-        background: #f1f1f1;
-        border-radius: 4px;
-    }
-    .document-viewer::-webkit-scrollbar-thumb {
-        background: #888;
-        border-radius: 4px;
-    }
-    .document-viewer::-webkit-scrollbar-thumb:hover {
-        background: #555;
-    }
-    .current-paragraph-highlight {
-        background: linear-gradient(120deg, #e3f2fd, #bbdefb);
-        padding: 8px 12px;
-        margin: 8px 0;
-        border-left: 4px solid #2196f3;
-        border-radius: 4px;
-        box-shadow: 0 2px 4px rgba(33, 150, 243, 0.2);
-    }
-    .reading-content {
-        margin-bottom: 10px;
-    }
-    .controls-section {
-        margin-top: 15px;
-    }
-    /* Reduce sidebar header spacing */
-    .sidebar .sidebar-content {
-        padding-top: 1rem;
-    }
-    /* FIXED: Make expander headers consistent with sidebar titles */
-    .streamlit-expanderHeader {
-        font-size: 1.1rem !important;
-        font-weight: 600 !important;
-        color: inherit !important;
-    }
-    /* FIXED: Ensure consistent styling for smart splitting expander */
-    div[data-testid="stExpander"] details summary p {
-        font-size: 1.1rem !important;
-        font-weight: 600 !important;
-    }
-    /* Style for the smart splitting section specifically */
-    .smart-splitting-header {
-        font-size: 1.1rem !important;
-        font-weight: 600 !important;
-    }
-    </style>
-    """, unsafe_allow_html=True)
-    
-    # Main title with LARGER font and MAXIMUM REDUCED TOP SPACE
-    st.markdown('<div class="main-title">🎙️Taqbaylit TTS Sɣer Adlis</div>', unsafe_allow_html=True)
-
-    if model is None:
-        st.warning("⚠️ Using test audio (TTS model not available)")
-
-    # Process audio queue on every run
-    if st.session_state.get('background_worker_started', False):
-        processed_count = process_audio_queue(st.session_state.audio_queue)
-        if processed_count > 0:
-            st.success(f"📥 Loaded {processed_count} paragraph(s)")
-
-    # Sidebar with KABYLE TRANSLATIONS
-    with st.sidebar:
-        # Document Settings with reduced spacing
-        st.header("📁 Tawila n Tɣuri", anchor=False)
-        uploaded_file = st.file_uploader("Sali-d Aḍris - jbed sers afaylu", type=['pdf', 'txt'], 
-                  help="Ulac OCR ara yerren afaylu PDF n tugna ɣer txt deg teqaylit. Ma ur iwulem ara ɛiwed-as tamuɣli.")
-
-        # Clear document button - MOVED UP immediately under file uploader
-        if st.session_state.get('processed'):
-            if st.button("🗑️ Sfeḍ Afaylu-a", type="secondary", use_container_width=True):
-                # Reset all document-related session state
-                st.session_state.processed = False
-                st.session_state.current_paragraph_index = 0
-                st.session_state.total_paragraphs = 0
-                st.session_state.paragraphs_data = {}
-                st.session_state.audio_ready = False
-                st.session_state.background_worker_started = False
-                st.session_state.reading_status = "Ready to start reading"
-                st.session_state.current_document_id = None
-                st.session_state.audio_queue = queue.Queue()
-                st.session_state.full_document_text = ""
-                st.rerun()
-        
-        # Playback Speed Control
-        st.markdown("---")
-        st.markdown("### 🎵 Arured n Tɣuri")
-        
-        # Define the speed options with labels
-        speed_options = [0.5, 0.8, 0.9, 1.0, 1.1, 1.2, 1.3, 1.5]
-        speed_labels = {
-            0.5: "0.5x (Ddac ddac ugar)",
-            0.8: "0.8x (Ddac ddac)", 
-            0.9: "0.9x (Ddac ddac cwiṭ)",
-            1.0: "1.0x (Amagnu)", 
-            1.1: "1.1x (Ɣiwel cwiṭ)",
-            1.2: "1.2x (Ɣiwel)",
-            1.3: "1.3x (Ɣiwel aṭas)",
-            1.5: "1.5x (Ɣiwel aṭas ugar)"
-        }
-        
-        # Create a select slider for playback speed
-        current_speed = st.select_slider(
-            "Asenfel n urured n tɣuri",
-            options=speed_options,
-            value=st.session_state.playback_speed,
-            format_func=lambda x: speed_labels[x],
-            help="Senfel arured n tɣuri i yal taseddaṛt"
-        )
-        
-        # Update session state if speed changed
-        if current_speed != st.session_state.playback_speed:
-            st.session_state.playback_speed = current_speed
-           # st.success(f"🎵 Arured n tɣuri yettwasenfel ɣer {speed_labels[current_speed]}")
-        
-        # Show current speed effect
-        speed_effect = {
-            0.5: "⏳ Ugar n ukud i tmesliwt",
-            0.8: "🐢 Fessus i uḍfaṛ",
-            0.9: "📝 S ttawil i usishel n tigzi", 
-            1.0: "✅ Arured amagnu n tmeslayt",
-            1.1: "⚡ Taɣuri s cwiṭ n uɣiwel",
-            1.2: "🚀 Taɣuri s uɣiwel",
-            1.3: "🎯 Taɣuri s uɣiwel d tmellit",
-            1.5: "🔥 Taɣuri s uɣiwel yuzzlen - i yimazzayen"
-        }
-        st.caption(f"**Asemdu:** {speed_effect[current_speed]}")
-        
-        # Smart splitting configuration - FIXED FONT STYLING
-        st.markdown("---")
-        
-        # Collapsible section for Smart Splitting with PROPER FONT HARMONIZATION
-        with st.expander("🎯 Agzam Amegzu", expanded=st.session_state.smart_splitting_expanded):
-            initial_paragraphs = st.slider("Tiseddaṛin deg tazwara", 3, 10, 5, 
-                                          help="Amḍan n tseddaṛin wezzilen deg tazwara n tɣuri")
-            initial_word_target = st.slider("Awalen deg tseddaṛt n tazwara", 30, 100, 50,
-                                           help="Amḍan n wawalen deg tseddaṛin n tazwara")
-            normal_word_target = st.slider("Awalen deg tseddaṛt tamagnut", 50, 300, 100,
-                                          help="Amḍan n wawalen deg tseddaṛin tineggura")
-
-    # Main content
-    # col_doc, col_reading = st.columns([2, 3])  # 40% document, 60% reading
-    col_reading, col_doc = st.columns([3, 2])  # 60% reading, 40% document
-
-    with col_doc:
-        # Kabyle title for document overview with SMALLER font
-        st.markdown('<div class="section-title">📄 Tamuɣli Ɣer Uḍris</div>', unsafe_allow_html=True)
-        
-        if st.session_state.get('full_document_text'):
-            # Display the full document in a scrollable container
-            document_html = "<div class='document-viewer'>"
-            paragraphs = st.session_state.get('paragraphs_list', [])
-            current_index = st.session_state.current_paragraph_index
-            
-            for i, paragraph in enumerate(paragraphs):
-                if i == current_index:
-                    document_html += f"<div class='current-paragraph-highlight'><strong>📍 Taseddaṛt {i+1}</strong><br>{paragraph}</div>"
-                else:
-                    document_html += f"<div><strong>Taseddaṛt {i+1}</strong><br>{paragraph}</div>"
-                if i < len(paragraphs) - 1:
-                    document_html += "<hr style='margin: 10px 0;'>"
-            
-            document_html += "</div>"
-            st.markdown(document_html, unsafe_allow_html=True)
-            
-            st.caption(f"📋 Aḍris: {len(paragraphs)} n tseddarin | 📍 Imir-a: Taseddaṛt {current_index + 1}")
-        else:
-            st.info("📁 Sali-d afaylu akken ad d-ikad da")
-            if uploaded_file and st.session_state.get('processed'):
-                st.warning("⚠️ Aḍris yettwasleḍ maca ulac-it. Ttxil-k, ɛiwed-as tikelt-nniḍen.")
-
-    with col_reading:
-        # Kabyle title for audio reading with SMALLER font
-        st.markdown('<div class="section-title">🎵 Ɣer - Sel : Aḍris</div>', unsafe_allow_html=True)
-        
-        if st.session_state.get('processed') and st.session_state.get('audio_ready', False):
-            current_index = st.session_state.current_paragraph_index
-            
-            # Check if current paragraph data exists
-            if current_index not in st.session_state.paragraphs_data:
-                st.warning(f"⏳ Taseddaṛt {current_index + 1} mazal d-tuli...")
-                st.info("Ameslaw n tseddaṛt-a mazal yemmid. Rǧu cwiṭ.")
-                # Auto-refresh to check for new data
-                time.sleep(2)
-                st.rerun()
-                return
-            
-            current_data = st.session_state.paragraphs_data[current_index]
-            
-            # NAVIGATION LAYOUT: Back + Audio + Next in one row
-            col_back, col_audio, col_next = st.columns([1, 2, 1])
-            
-            with col_back:
-                # BACK BUTTON
-                if current_index > 0:
-                    if st.button("⏮️ Deffir", 
-                                use_container_width=True,
-                                type="secondary",
-                                key=f"back_btn_top_{current_index}"):
-                        prev_index = current_index - 1
-                        st.session_state.current_paragraph_index = prev_index
-                        st.session_state.reading_status = f"Taɣuri n tseddaṛt {prev_index + 1}/{st.session_state.total_paragraphs}"
-                        st.rerun()
-                else:
-                    st.button("⏮️ Deffir", disabled=True, use_container_width=True)
-
-            with col_audio:
-                # Audio player status placeholder - will be in the HTML
-                pass
-                
-            with col_next:
-                # NEXT BUTTON
-                if current_index < st.session_state.total_paragraphs - 1:
-                    next_index = current_index + 1
-                    
-                    if st.button("⏭️ Sdat", 
-                               type="primary",
-                               use_container_width=True,
-                               key=f"next_btn_top_{current_index}"):
-                        
-                        st.session_state.current_paragraph_index = next_index
-                        st.session_state.reading_status = f"Taɣuri n tseddaṛt {next_index + 1}/{st.session_state.total_paragraphs}"
-                        st.rerun()
-                    
-                    # Status indicator (informational only)
-                    next_ready = next_index in st.session_state.paragraphs_data
-                    ready_count = len(st.session_state.paragraphs_data)
-                    total_count = st.session_state.total_paragraphs
-                    if not next_ready:
-                        st.caption(f"⏳ Yesselkat... ({ready_count}/{total_count})")
-                    else:
-                        st.caption(f"✅ Yemmed ({ready_count}/{total_count})")
-                        
-                else:
-                    st.button("⏭️ Sdat", disabled=True, use_container_width=True)
-                    st.caption("🎉 Temmed tɣuri!")
-
-            # HTML with real-time text highlighting
-            timing_json = json.dumps(current_data['timing_info'])
-            full_text = current_data['paragraph_text'].replace('`', '\\`').replace('${', '\\${')
-            audio_b64 = base64.b64encode(current_data['audio_bytes']).decode()
-            playback_speed = st.session_state.playback_speed
-            
-            complete_html = f"""
-            <!DOCTYPE html>
-            <html>
-            <head>
-                <style>
-                .phrase-highlight {{
-                    background: linear-gradient(120deg, #ffeb3b, #ffd54f);
-                    padding: 4px 8px;
-                    margin: 2px 1px;
-                    border-radius: 6px;
-                    box-shadow: 0 2px 4px rgba(255, 235, 59, 0.3);
-                    font-weight: bold;
-                    transition: all 0.3s ease;
-                }}
-                .phrase-text {{
-                    padding: 2px 4px;
-                    margin: 1px 0px;
-                    border-radius: 4px;
-                    transition: all 0.3s ease;
-                }}
-                .reading-container {{
-                    max-height: 40vh;
-                    overflow-y: auto;
-                    padding: 20px;
-                    border: 2px solid #e1e1e1;
-                    border-radius: 12px;
-                    background: #fafafa;
-                    margin: 5px 0;
-                    line-height: 1.8;
-                    font-size: 16px;
-                    font-family: Arial, sans-serif;
-                }}
-                .status-bar {{
-                    background: #e3f2fd;
-                    padding: 8px;
-                    border-radius: 8px;
-                    margin: 8px 0;
-                    font-size: 14px;
-                }}
-                .audio-player {{
-                    width: 100%;
-                    margin: 8px 0;
-                    text-align: center;
-                }}
-                .audio-controls {{
-                    display: flex;
-                    justify-content: center;
-                    align-items: center;
-                    gap: 10px;
-                    margin-bottom: 10px;
-                }}
-                </style>
-            </head>
-            <body>
-                <div class="audio-player">
-                    <div class="audio-controls">
-                        <audio id="mainAudio" controls playbackRate={playback_speed} style="min-width: 250px;">
-                            <source src="data:audio/wav;base64,{audio_b64}" type="audio/wav">
-                        </audio>
-                    </div>
-                    <div style="margin-top: 5px; font-size: 14px; color: #666;">
-                        🎵 Seɣbel, tekki ɣef ▶️ afella i tɣuri s urured: <strong>{playback_speed}x</strong>
-                        {"" if playback_speed == 1.0 else " - " + ("ddac ddac" if playback_speed < 1.0 else "aɣiwel")}
-                    </div>
-                </div>
-                
-
-                
-                <div class="reading-container" id="readingContainer">
-                    {full_text}
-                </div>
-
-                <script>
-                const timingInfo = {timing_json};
-                const fullText = `{full_text}`;
-                const playbackSpeed = {playback_speed};
-                let currentHighlightIndex = -1;
-                let phraseElements = [];
-                
-                function initializeHighlighting() {{
-                    // Create phrase elements by wrapping text
-                    let container = document.getElementById('readingContainer');
-                    let workingText = fullText;
-                    
-                    timingInfo.forEach((phrase, index) => {{
-                        const cleanPhrase = phrase.text.trim();
-                        if (workingText.includes(cleanPhrase)) {{
-                            const spanId = 'phrase_' + index;
-                            const spanHtml = '<span id="' + spanId + '" class="phrase-text">' + cleanPhrase + '</span>';
-                            workingText = workingText.replace(cleanPhrase, spanHtml);
-                        }}
-                    }});
-                    
-                    container.innerHTML = workingText;
-                    
-                    // Store references to all phrase elements
-                    timingInfo.forEach((phrase, index) => {{
-                        const element = document.getElementById('phrase_' + index);
-                        if (element) {{
-                            phraseElements.push(element);
-                        }}
-                    }});
-                    
-                    updateDebugInfo("Agzam amegzu yemmed! " + phraseElements.length + " n tefyar s " + playbackSpeed + "x arured");
-                }}
-                
-                function updateDebugInfo(message) {{
-                    const debugEl = document.getElementById('debugInfo');
-                    if (debugEl) debugEl.textContent = message;
-                }}
-                
-                function highlightCurrentPhrase(currentTime) {{
-                    let newIndex = -1;
-                    for (let i = 0; i < timingInfo.length; i++) {{
-                        if (currentTime >= timingInfo[i].start && currentTime < timingInfo[i].end) {{
-                            newIndex = i;
-                            break;
-                        }}
-                    }}
-                    
-                    if (newIndex !== currentHighlightIndex) {{
-                        currentHighlightIndex = newIndex;
-                        updateHighlightDisplay();
-                        if (newIndex >= 0) {{
-                            updateDebugInfo("Akud: " + currentTime.toFixed(2) + "s | Tafyirt: " + (newIndex + 1) + "/" + timingInfo.length + " | Arured: " + playbackSpeed + "x");
-                        }}
-                    }}
-                }}
-                
-                function updateHighlightDisplay() {{
-                    const currentPhraseSpan = document.getElementById('currentPhrase');
-                    
-                    // Remove all highlights
-                    phraseElements.forEach(element => {{
-                        element.className = 'phrase-text';
-                    }});
-                    
-                    // Highlight current phrase
-                    if (currentHighlightIndex >= 0 && currentHighlightIndex < phraseElements.length) {{
-                        const element = phraseElements[currentHighlightIndex];
-                        if (element) {{
-                            element.className = 'phrase-highlight';
-                            element.scrollIntoView({{ behavior: 'smooth', block: 'center' }});
-                        }}
-                        
-                        if (currentPhraseSpan && timingInfo[currentHighlightIndex]) {{
-                            currentPhraseSpan.textContent = timingInfo[currentHighlightIndex].text.substring(0, 100) + 
-                                (timingInfo[currentHighlightIndex].text.length > 100 ? '...' : '');
-                        }}
-                    }} else {{
-                        if (currentPhraseSpan) {{
-                            currentPhraseSpan.textContent = 'Araǧu amesli...';
-                        }}
-                    }}
-                }}
-                
-                // Set up audio event listeners
-                function setupAudioListeners() {{
-                    const audioElement = document.getElementById('mainAudio');
-                    if (audioElement) {{
-                        // Set playback rate
-                        audioElement.playbackRate = playbackSpeed;
-                        
-                        audioElement.addEventListener('timeupdate', function() {{
-                            highlightCurrentPhrase(this.currentTime);
-                        }});
-                        
-                        audioElement.addEventListener('play', function() {{
-                            updateDebugInfo("🎵 Taɣuri... aseḍfeṛ n tira iteddu s " + playbackSpeed + "x arured");
-                        }});
-                        
-                        audioElement.addEventListener('ended', function() {{
-                            currentHighlightIndex = -1;
-                            updateHighlightDisplay();
-                            updateDebugInfo("✅ Taɣuri tekfa s " + playbackSpeed + "x arured");
-                        }});
-                        
-                    }} else {{
-                        setTimeout(setupAudioListeners, 100);
-                    }}
-                }}
-                
-                // Initialize everything when page loads
-                document.addEventListener('DOMContentLoaded', function() {{
-                    initializeHighlighting();
-                    setupAudioListeners();
-                }});
-                </script>
-            </body>
-            </html>
-            """
-            
-            # Display the complete reading content
-            st.components.v1.html(complete_html, height=300, scrolling=True)
-            
-            # Place the remaining controls BELOW the reading content
-            st.markdown('<div class="controls-section">', unsafe_allow_html=True)
-            
-            # Show paragraph info
-            word_count = current_data.get('word_count', len(current_data['paragraph_text'].split()))
-            st.markdown(f"**Taseddaṛt {current_index + 1}/{st.session_state.total_paragraphs}**")
-            st.caption(f"📊 {word_count} n wawalen | ⏱️ {current_data['audio_duration']:.1f}s | 🎵 {st.session_state.playback_speed}x arured")
-
-            # Display progress
-            ready_count = len(st.session_state.paragraphs_data)
-            total_count = st.session_state.total_paragraphs
-            progress = ready_count / total_count if total_count > 0 else 0
-            st.progress(progress)
-            st.caption(f"📊 Asekker: {ready_count}/{total_count} n tseddarin mmedent ({progress:.0%})")
-            
-            # Download button for current paragraph
-            audio_bytes = current_data['audio_bytes']
-            st.download_button(
-                "📥 Zdem ameslaw n tseddaṛt-a",
-                audio_bytes,
-                f"Taseddaṛt_{current_index + 1}.wav",
-                "audio/wav",
-                use_container_width=True
-            )
-            
-            st.markdown('</div>', unsafe_allow_html=True)
-
-        elif uploaded_file and not st.session_state.processed:
-            if st.button("🔄 Selket Aḍris", type="primary"):
-                # Process document when button is clicked
-                with st.spinner("Asekker n uḍris s ugzam amegzu n tseddaṛin..."):
-                    temp_path = os.path.join(tempfile.gettempdir(), uploaded_file.name)
-                    with open(temp_path, "wb") as f:
-                        f.write(uploaded_file.getvalue())
-
-                    text, error = read_document(temp_path)
-                    if error:
-                        st.error(error)
-                    else:
-                        cleaned_text = clean_text(text)
-                        st.session_state.full_document_text = cleaned_text
-                        
-                        # Use smart splitting strategy
-                        paragraphs = smart_split_paragraphs(
-                            cleaned_text, 
-                            initial_paragraphs=initial_paragraphs,
-                            initial_word_target=initial_word_target,
-                            normal_word_target=normal_word_target
-                        )
-                        
-                        if not paragraphs:
-                            st.error("Ulac agbur i tɣuri.")
-                            return
-                        
-                        # Initialize processing state
-                        st.session_state.total_paragraphs = len(paragraphs)
-                        st.session_state.current_paragraph_index = 0
-                        st.session_state.paragraphs_data = {}
-                        st.session_state.paragraphs_list = paragraphs
-                        st.session_state.processed = True
-                        
-                        # Generate first paragraph immediately in main thread
-                        first_paragraph = paragraphs[0]
-                        audio_data, sampling_rate, timing_info = generate_audio_with_precise_timing(first_paragraph)
-                        
-                        st.session_state.paragraphs_data[0] = {
-                            'paragraph_text': first_paragraph,
-                            'audio_data': audio_data,
-                            'timing_info': timing_info,
-                            'audio_duration': len(audio_data) / sampling_rate,
-                            'audio_bytes': get_audio_bytes(audio_data),
-                            'word_count': len(first_paragraph.split())
-                        }
-                        st.session_state.audio_ready = True
-                        
-                        # Start background worker for ALL remaining paragraphs
-                        if len(paragraphs) > 1:
-                            remaining_paragraphs = paragraphs[1:]
-                            
-                            # Use queue-based background worker
-                            thread = threading.Thread(
-                                target=background_audio_worker,
-                                args=(remaining_paragraphs, st.session_state.audio_queue, 1)
-                            )
-                            thread.daemon = True
-                            thread.start()
-                            
-                            st.session_state.background_worker_started = True
-                        
-                        st.rerun()
-            else:
-                st.info("🔄 Seɣbel, tekki ɣef 'Selket Aḍris' iwakken ad yettwasleḍ u ad yeddu seg tira ɣer umeslaw")
-        else:
-            st.info("👆 Sali-d afaylu iwakken ad tedduḍ ɣer tɣuri")
-
-if __name__ == "__main__":
+# Specific conda environment : Kabyle_streamlit
+
+# Streamlit TTS Reader - Taqbaylit TTS Sɣer Adlis
+# Bouaziz Ait Driss October 2025
+
+import streamlit as st
+import fitz  # PyMuPDF
+import re
+import numpy as np
+import torch
+from scipy.io.wavfile import write as wav_write
+import tempfile
+import os
+import base64
+import json
+import io
+import threading
+import queue
+import time
+import pdfplumber
+
+# Configure Streamlit page
+st.set_page_config(
+    page_title="Kabyle TTS Document Reader",
+    page_icon="🎙️",
+    layout="wide",
+    initial_sidebar_state="expanded",
+)
+
+# Try to import transformers
+try:
+    from transformers import VitsModel, AutoTokenizer
+    TRANSFORMERS_AVAILABLE = True
+except ImportError:
+    try:
+        from transformers import AutoModel, AutoTokenizer
+        TRANSFORMERS_AVAILABLE = True
+        VitsModel = AutoModel
+    except ImportError:
+        TRANSFORMERS_AVAILABLE = False
+
+# --- Model Loading ---
+@st.cache_resource
+def load_model():
+    if not TRANSFORMERS_AVAILABLE:
+        return None, None, "cpu"
+    try:
+        model = VitsModel.from_pretrained("facebook/mms-tts-kab")
+        tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-kab")
+        device = "cuda" if torch.cuda.is_available() else "cpu"
+        model = model.to(device)
+        return model, tokenizer, device
+    except Exception as e:
+        st.error(f"Error loading model: {e}")
+        return None, None, "cpu"
+
+model, tokenizer, device = load_model()
+
+# --- Text Processing ---
+def clean_text(text):
+    if not text:
+        return text
+    text = re.sub(r' +', ' ', text)
+    text = re.sub(r'\n\s*\n', '\n\n', text)
+    return text.strip()
+
+def smart_split_paragraphs(text, initial_paragraphs=5, initial_word_target=50, normal_word_target=200):
+    """
+    Smart paragraph splitting strategy:
+    - First 'initial_paragraphs' paragraphs: ~initial_word_target words each (for quick startup)
+    - Remaining paragraphs: ~normal_word_target words each (for better timing)
+    """
+    if not text:
+        return []
+    
+    # First, split by major paragraph breaks
+    raw_paragraphs = re.split(r'\n\s*\n', text)
+    raw_paragraphs = [p.strip() for p in raw_paragraphs if p.strip()]
+    
+    if not raw_paragraphs:
+        return []
+    
+    final_paragraphs = []
+    current_sentences = []
+    
+    # Split each raw paragraph into sentences
+    all_sentences = []
+    for paragraph in raw_paragraphs:
+        # Split on sentence endings: . ! ? : ; followed by space
+        sentences = re.split(r'(?<=[.!?:;])\s+', paragraph)
+        sentences = [s.strip() for s in sentences if s.strip()]
+        all_sentences.extend(sentences)
+    
+    if not all_sentences:
+        return []
+    
+    # Build initial quick-start paragraphs (shorter)
+    word_count = 0
+    for sentence in all_sentences:
+        sentence_word_count = len(sentence.split())
+        
+        # For first few paragraphs, use smaller target
+        if len(final_paragraphs) < initial_paragraphs:
+            target_word_count = initial_word_target
+        else:
+            target_word_count = normal_word_target
+        
+        # If adding this sentence would exceed target and we have some content, start new paragraph
+        if current_sentences and word_count + sentence_word_count > target_word_count:
+            # Join current sentences to form a paragraph
+            paragraph_text = ' '.join(current_sentences)
+            final_paragraphs.append(paragraph_text)
+            current_sentences = [sentence]
+            word_count = sentence_word_count
+        else:
+            current_sentences.append(sentence)
+            word_count += sentence_word_count
+    
+    # Add the last paragraph if we have remaining sentences
+    if current_sentences:
+        paragraph_text = ' '.join(current_sentences)
+        final_paragraphs.append(paragraph_text)
+    
+    return final_paragraphs
+
+def split_paragraph_into_phrases(paragraph, max_phrase_length=150):
+    if not paragraph:
+        return []
+    
+    phrases = []
+    current_phrase = ""
+    sentences = re.split(r'(?<=[.!?])\s+', paragraph)
+    
+    for sentence in sentences:
+        if len(current_phrase) + len(sentence) > max_phrase_length and current_phrase:
+            phrases.append(current_phrase.strip())
+            current_phrase = sentence
+        else:
+            current_phrase = current_phrase + " " + sentence if current_phrase else sentence
+
+    if current_phrase.strip():
+        phrases.append(current_phrase.strip())
+
+    return phrases
+
+def generate_audio_with_precise_timing(paragraph):
+    """Generate audio with precise phrase-level timing using per-phrase generation"""
+    
+    if model is None or tokenizer is None:
+        return create_dummy_audio_for_paragraph(paragraph), 16000, []
+
+    try:
+        phrases = split_paragraph_into_phrases(paragraph)
+        if not phrases:
+            return create_dummy_audio_for_paragraph(paragraph), 16000, []
+
+        all_audio = []
+        phrase_durations = []
+        sampling_rate = 16000
+        
+        # Generate audio for each phrase separately to get exact timing
+        for i, phrase in enumerate(phrases):
+            if not phrase.strip():
+                continue
+                
+            # Generate audio for this specific phrase
+            inputs = tokenizer(phrase, return_tensors="pt").to(device)
+            with torch.no_grad():
+                output = model(**inputs)
+            
+            # Extract audio data
+            if hasattr(output, 'waveform'):
+                audio_chunk = output.waveform.cpu().numpy().squeeze()
+            else:
+                audio_chunk = output[0].cpu().numpy().squeeze() if len(output) > 0 else None
+            
+            if audio_chunk is not None:
+                # Calculate exact duration for this phrase
+                phrase_duration = len(audio_chunk) / sampling_rate
+                phrase_durations.append(phrase_duration)
+                
+                # Normalize and store
+                max_val = np.max(np.abs(audio_chunk))
+                if max_val > 0:
+                    audio_chunk = audio_chunk / max_val * 0.8
+                all_audio.append(audio_chunk)
+            else:
+                # Fallback: estimate duration if audio generation failed
+                word_count = len(phrase.split())
+                estimated_duration = word_count * 0.4
+                phrase_durations.append(estimated_duration)
+
+        if all_audio:
+            # Concatenate all phrase audio
+            full_audio = np.concatenate(all_audio)
+            total_duration = len(full_audio) / sampling_rate
+            
+            # Build precise timing info using actual phrase durations
+            timing_info = []
+            current_time = 0.0
+            
+            for i, (phrase, duration) in enumerate(zip(phrases, phrase_durations)):
+                timing_info.append({
+                    'text': phrase,
+                    'start': current_time,
+                    'end': current_time + duration,
+                    'duration': duration
+                })
+                current_time += duration
+            
+            return full_audio, sampling_rate, timing_info
+        else:
+            return create_dummy_audio_for_paragraph(paragraph), 16000, []
+            
+    except Exception as e:
+        return create_dummy_audio_for_paragraph(paragraph), 16000, []
+
+def create_dummy_audio_for_paragraph(paragraph):
+    word_count = len(paragraph.split())
+    total_duration = word_count * 0.4
+    sampling_rate = 16000
+    t = np.linspace(0, total_duration, int(sampling_rate * total_duration))
+    audio = 0.3 * np.sin(2 * np.pi * 220 * t)
+    return audio
+
+# Add a fallback to plumber for reading diacritic letters
+def read_document(file_path):
+    text = ""
+    try:
+        if file_path.lower().endswith('.pdf'):
+            # FIRST TRY: PyMuPDF (faster for digital PDFs)
+            try:
+                pdf_document = fitz.open(file_path)
+                for page_num in range(len(pdf_document)):
+                    page = pdf_document[page_num]
+                    text += page.get_text() + "\n"
+                pdf_document.close()
+                
+                # Check if we got meaningful text with diacritics
+                if text.strip() and any(char in text for char in ['é', 'è', 'à', 'ù', 'ï', 'ö', 'α', 'β', 'γ']):
+                    return text, []
+                    
+            except Exception as e:
+                text = ""  # Reset text if PyMuPDF fails
+                
+            # FALLBACK: pdfplumber (better for scanned/diacritic PDFs)
+            try:
+                import pdfplumber
+                with pdfplumber.open(file_path) as pdf:
+                    text = ""
+                    for page in pdf.pages:
+                        page_text = page.extract_text() or ""
+                        text += page_text + "\n"
+            except ImportError:
+                return "pdfplumber not available", []
+            except Exception as e:
+                return f"Both PDF methods failed: {e}", []
+                
+        elif file_path.lower().endswith('.txt'):
+            with open(file_path, 'r', encoding='utf-8') as f:
+                text = f.read()
+        else:
+            return "Unsupported file format", []
+            
+    except Exception as e:
+        return f"Error reading document: {e}", []
+        
+    return text, []
+
+def normalize_audio(audio_data):
+    if audio_data.dtype != np.float32:
+        audio_data = audio_data.astype(np.float32)
+    max_val = np.max(np.abs(audio_data))
+    if max_val > 0:
+        audio_data = audio_data / max_val * 0.9
+    return audio_data
+
+def get_audio_bytes(audio_data):
+    audio_bytes = io.BytesIO()
+    audio_data = normalize_audio(audio_data)
+    audio_int16 = np.clip(audio_data * 32767, -32768, 32767).astype(np.int16)
+    wav_write(audio_bytes, 16000, audio_int16)
+    audio_bytes.seek(0)
+    return audio_bytes.getvalue()
+
+# --- Background Audio Generation ---
+def background_audio_worker(paragraphs, audio_queue, start_index=0):
+    """Background worker to generate audio for paragraphs - NO SESSION STATE ACCESS"""
+    try:
+        for i, paragraph in enumerate(paragraphs):
+            paragraph_index = start_index + i
+            word_count = len(paragraph.split())
+            
+            # Generate audio with precise timing
+            audio_data, sampling_rate, timing_info = generate_audio_with_precise_timing(paragraph)
+            
+            # Send to main thread via queue ONLY
+            audio_queue.put({
+                'paragraph_index': paragraph_index,
+                'paragraph_text': paragraph,
+                'audio_data': audio_data,
+                'timing_info': timing_info,
+                'audio_duration': len(audio_data) / sampling_rate,
+                'audio_bytes': get_audio_bytes(audio_data),
+                'word_count': word_count
+            })
+            
+    except Exception as e:
+        pass
+
+# --- Queue Processing ---
+def process_audio_queue(audio_queue):
+    """Process all available items in the audio queue - called from main thread only"""
+    processed_count = 0
+    try:
+        while True:
+            # Non-blocking check
+            audio_data = audio_queue.get_nowait()
+            
+            # SAFE: Main thread updating session state
+            st.session_state.paragraphs_data[audio_data['paragraph_index']] = audio_data
+            processed_count += 1
+            
+    except queue.Empty:
+        pass
+    
+    return processed_count
+
+# Initialize session state
+if 'processed' not in st.session_state:
+    st.session_state.processed = False
+if 'current_paragraph_index' not in st.session_state:
+    st.session_state.current_paragraph_index = 0
+if 'total_paragraphs' not in st.session_state:
+    st.session_state.total_paragraphs = 0
+if 'paragraphs_data' not in st.session_state:
+    st.session_state.paragraphs_data = {}
+if 'audio_ready' not in st.session_state:
+    st.session_state.audio_ready = False
+if 'background_worker_started' not in st.session_state:
+    st.session_state.background_worker_started = False
+if 'reading_status' not in st.session_state:
+    st.session_state.reading_status = "Ready to start reading"
+if 'current_document_id' not in st.session_state:
+    st.session_state.current_document_id = None
+if 'audio_queue' not in st.session_state:
+    st.session_state.audio_queue = queue.Queue()
+if 'playback_speed' not in st.session_state:
+    st.session_state.playback_speed = 1.0  # Default normal speed
+if 'full_document_text' not in st.session_state:
+    st.session_state.full_document_text = ""
+if 'smart_splitting_expanded' not in st.session_state:
+    st.session_state.smart_splitting_expanded = False
+
+# --- Streamlit UI ---
+def main():
+    # Custom CSS for better styling - FIXED FONT HARMONIZATION
+    st.markdown("""
+    <style>
+    .main-title {
+        font-size: 1.5rem !important;
+        font-weight: bold !important;
+        margin-bottom: -1rem !important;
+        margin-top: -1rem !important;      /* MAXIMUM REDUCED top margin */
+        color: #1f77b4;
+    }
+    .section-title {
+        font-size: 1.3rem !important;
+        font-weight: bold !important;
+        margin-bottom: 0.3rem !important;
+        margin-top: 0rem !important;
+        color: #2e86ab;
+    }
+    .document-viewer {
+        max-height: 70vh;
+        overflow-y: auto;
+        border: 1px solid #e1e1e1;
+        border-radius: 8px;
+        padding: 15px;
+        background: #fafafa;
+        font-family: Arial, sans-serif;
+        line-height: 1.6;
+        scrollbar-width: thin;
+        scrollbar-color: #888 #f1f1f1;
+        color: #333333;  /* ADD THIS LINE - ensures dark text */
+    }
+    .document-viewer::-webkit-scrollbar {
+        width: 8px;
+    }
+    .document-viewer::-webkit-scrollbar-track {
+        background: #f1f1f1;
+        border-radius: 4px;
+    }
+    .document-viewer::-webkit-scrollbar-thumb {
+        background: #888;
+        border-radius: 4px;
+    }
+    .document-viewer::-webkit-scrollbar-thumb:hover {
+        background: #555;
+    }
+    .current-paragraph-highlight {
+        background: linear-gradient(120deg, #e3f2fd, #bbdefb);
+        padding: 8px 12px;
+        margin: 8px 0;
+        border-left: 4px solid #2196f3;
+        border-radius: 4px;
+        box-shadow: 0 2px 4px rgba(33, 150, 243, 0.2);
+    }
+    .reading-content {
+        margin-bottom: 10px;
+    }
+    .controls-section {
+        margin-top: 15px;
+    }
+    /* Reduce sidebar header spacing */
+    .sidebar .sidebar-content {
+        padding-top: 1rem;
+    }
+    /* FIXED: Make expander headers consistent with sidebar titles */
+    .streamlit-expanderHeader {
+        font-size: 1.1rem !important;
+        font-weight: 600 !important;
+        color: inherit !important;
+    }
+    /* FIXED: Ensure consistent styling for smart splitting expander */
+    div[data-testid="stExpander"] details summary p {
+        font-size: 1.1rem !important;
+        font-weight: 600 !important;
+    }
+    /* Style for the smart splitting section specifically */
+    .smart-splitting-header {
+        font-size: 1.1rem !important;
+        font-weight: 600 !important;
+    }
+    </style>
+    """, unsafe_allow_html=True)
+    
+    # Main title with LARGER font and MAXIMUM REDUCED TOP SPACE
+    st.markdown('<div class="main-title">🎙️Taqbaylit TTS Sɣer Adlis</div>', unsafe_allow_html=True)
+
+    if model is None:
+        st.warning("⚠️ Using test audio (TTS model not available)")
+
+    # Process audio queue on every run
+    if st.session_state.get('background_worker_started', False):
+        processed_count = process_audio_queue(st.session_state.audio_queue)
+        if processed_count > 0:
+            st.success(f"📥 Loaded {processed_count} paragraph(s)")
+
+    # Sidebar with KABYLE TRANSLATIONS
+    with st.sidebar:
+        # Document Settings with reduced spacing
+        st.header("📁 Tawila n Tɣuri", anchor=False)
+        uploaded_file = st.file_uploader("Sali-d Aḍris - jbed sers afaylu", type=['pdf', 'txt'], 
+                  help="Ulac OCR ara yerren afaylu PDF n tugna ɣer txt deg teqaylit. Ma ur iwulem ara ɛiwed-as tamuɣli.")
+
+        # Clear document button - MOVED UP immediately under file uploader
+        if st.session_state.get('processed'):
+            if st.button("🗑️ Sfeḍ Afaylu-a", type="secondary", use_container_width=True):
+                # Reset all document-related session state
+                st.session_state.processed = False
+                st.session_state.current_paragraph_index = 0
+                st.session_state.total_paragraphs = 0
+                st.session_state.paragraphs_data = {}
+                st.session_state.audio_ready = False
+                st.session_state.background_worker_started = False
+                st.session_state.reading_status = "Ready to start reading"
+                st.session_state.current_document_id = None
+                st.session_state.audio_queue = queue.Queue()
+                st.session_state.full_document_text = ""
+                st.rerun()
+        
+        # Playback Speed Control
+        st.markdown("---")
+        st.markdown("### 🎵 Arured n Tɣuri")
+        
+        # Define the speed options with labels
+        speed_options = [0.5, 0.8, 0.9, 1.0, 1.1, 1.2, 1.3, 1.5]
+        speed_labels = {
+            0.5: "0.5x (Ddac ddac ugar)",
+            0.8: "0.8x (Ddac ddac)", 
+            0.9: "0.9x (Ddac ddac cwiṭ)",
+            1.0: "1.0x (Amagnu)", 
+            1.1: "1.1x (Ɣiwel cwiṭ)",
+            1.2: "1.2x (Ɣiwel)",
+            1.3: "1.3x (Ɣiwel aṭas)",
+            1.5: "1.5x (Ɣiwel aṭas ugar)"
+        }
+        
+        # Create a select slider for playback speed
+        current_speed = st.select_slider(
+            "Asenfel n urured n tɣuri",
+            options=speed_options,
+            value=st.session_state.playback_speed,
+            format_func=lambda x: speed_labels[x],
+            help="Senfel arured n tɣuri i yal taseddaṛt"
+        )
+        
+        # Update session state if speed changed
+        if current_speed != st.session_state.playback_speed:
+            st.session_state.playback_speed = current_speed
+           # st.success(f"🎵 Arured n tɣuri yettwasenfel ɣer {speed_labels[current_speed]}")
+        
+        # Show current speed effect
+        speed_effect = {
+            0.5: "⏳ Ugar n ukud i tmesliwt",
+            0.8: "🐢 Fessus i uḍfaṛ",
+            0.9: "📝 S ttawil i usishel n tigzi", 
+            1.0: "✅ Arured amagnu n tmeslayt",
+            1.1: "⚡ Taɣuri s cwiṭ n uɣiwel",
+            1.2: "🚀 Taɣuri s uɣiwel",
+            1.3: "🎯 Taɣuri s uɣiwel d tmellit",
+            1.5: "🔥 Taɣuri s uɣiwel yuzzlen - i yimazzayen"
+        }
+        st.caption(f"**Asemdu:** {speed_effect[current_speed]}")
+        
+        # Smart splitting configuration - FIXED FONT STYLING
+        st.markdown("---")
+        
+        # Collapsible section for Smart Splitting with PROPER FONT HARMONIZATION
+        with st.expander("🎯 Agzam Amegzu", expanded=st.session_state.smart_splitting_expanded):
+            initial_paragraphs = st.slider("Tiseddaṛin deg tazwara", 3, 10, 5, 
+                                          help="Amḍan n tseddaṛin wezzilen deg tazwara n tɣuri")
+            initial_word_target = st.slider("Awalen deg tseddaṛt n tazwara", 30, 100, 50,
+                                           help="Amḍan n wawalen deg tseddaṛin n tazwara")
+            normal_word_target = st.slider("Awalen deg tseddaṛt tamagnut", 50, 300, 100,
+                                          help="Amḍan n wawalen deg tseddaṛin tineggura")
+
+    # Main content
+    # col_doc, col_reading = st.columns([2, 3])  # 40% document, 60% reading
+    col_reading, col_doc = st.columns([3, 2])  # 60% reading, 40% document
+
+    with col_doc:
+        # Kabyle title for document overview with SMALLER font
+        st.markdown('<div class="section-title">📄 Tamuɣli Ɣer Uḍris</div>', unsafe_allow_html=True)
+        
+        if st.session_state.get('full_document_text'):
+            # Display the full document in a scrollable container
+            document_html = "<div class='document-viewer'>"
+            paragraphs = st.session_state.get('paragraphs_list', [])
+            current_index = st.session_state.current_paragraph_index
+            
+            for i, paragraph in enumerate(paragraphs):
+                if i == current_index:
+                    document_html += f"<div class='current-paragraph-highlight'><strong>📍 Taseddaṛt {i+1}</strong><br>{paragraph}</div>"
+                else:
+                    document_html += f"<div><strong>Taseddaṛt {i+1}</strong><br>{paragraph}</div>"
+                if i < len(paragraphs) - 1:
+                    document_html += "<hr style='margin: 10px 0;'>"
+            
+            document_html += "</div>"
+            st.markdown(document_html, unsafe_allow_html=True)
+            
+            st.caption(f"📋 Aḍris: {len(paragraphs)} n tseddarin | 📍 Imir-a: Taseddaṛt {current_index + 1}")
+        else:
+            st.info("📁 Sali-d afaylu akken ad d-ikad da")
+            if uploaded_file and st.session_state.get('processed'):
+                st.warning("⚠️ Aḍris yettwasleḍ maca ulac-it. Ttxil-k, ɛiwed-as tikelt-nniḍen.")
+
+    with col_reading:
+        # Kabyle title for audio reading with SMALLER font
+        st.markdown('<div class="section-title">🎵 Ɣer - Sel  Aḍris</div>', unsafe_allow_html=True)
+        
+        if st.session_state.get('processed') and st.session_state.get('audio_ready', False):
+            current_index = st.session_state.current_paragraph_index
+            
+            # Check if current paragraph data exists
+            if current_index not in st.session_state.paragraphs_data:
+                st.warning(f"⏳ Taseddaṛt {current_index + 1} mazal d-tuli...")
+                st.info("Ameslaw n tseddaṛt-a mazal yemmid. Rǧu cwiṭ.")
+                # Auto-refresh to check for new data
+                time.sleep(2)
+                st.rerun()
+                return
+            
+            current_data = st.session_state.paragraphs_data[current_index]
+            
+            # NAVIGATION LAYOUT: Back + Audio + Next in one row
+            col_back, col_audio, col_next = st.columns([1, 2, 1])
+            
+            with col_back:
+                # BACK BUTTON
+                if current_index > 0:
+                    if st.button("⏮️ Deffir", 
+                                use_container_width=True,
+                                type="secondary",
+                                key=f"back_btn_top_{current_index}"):
+                        prev_index = current_index - 1
+                        st.session_state.current_paragraph_index = prev_index
+                        st.session_state.reading_status = f"Taɣuri n tseddaṛt {prev_index + 1}/{st.session_state.total_paragraphs}"
+                        st.rerun()
+                else:
+                    st.button("⏮️ Deffir", disabled=True, use_container_width=True)
+
+            with col_audio:
+                # Audio player status placeholder - will be in the HTML
+                pass
+                
+            with col_next:
+                # NEXT BUTTON
+                if current_index < st.session_state.total_paragraphs - 1:
+                    next_index = current_index + 1
+                    
+                    if st.button("⏭️ Sdat", 
+                               type="primary",
+                               use_container_width=True,
+                               key=f"next_btn_top_{current_index}"):
+                        
+                        st.session_state.current_paragraph_index = next_index
+                        st.session_state.reading_status = f"Taɣuri n tseddaṛt {next_index + 1}/{st.session_state.total_paragraphs}"
+                        st.rerun()
+                    
+                    # Status indicator (informational only)
+                    next_ready = next_index in st.session_state.paragraphs_data
+                    ready_count = len(st.session_state.paragraphs_data)
+                    total_count = st.session_state.total_paragraphs
+                    if not next_ready:
+                        st.caption(f"⏳ Yesselkat... ({ready_count}/{total_count})")
+                    else:
+                        st.caption(f"✅ Yemmed ({ready_count}/{total_count})")
+                        
+                else:
+                    st.button("⏭️ Sdat", disabled=True, use_container_width=True)
+                    st.caption("🎉 Temmed tɣuri!")
+
+            # HTML with real-time text highlighting
+            timing_json = json.dumps(current_data['timing_info'])
+            full_text = current_data['paragraph_text'].replace('`', '\\`').replace('${', '\\${')
+            audio_b64 = base64.b64encode(current_data['audio_bytes']).decode()
+            playback_speed = st.session_state.playback_speed
+            
+            complete_html = f"""
+            <!DOCTYPE html>
+            <html>
+            <head>
+                <style>
+                .phrase-highlight {{
+                    background: linear-gradient(120deg, #ffeb3b, #ffd54f);
+                    padding: 4px 8px;
+                    margin: 2px 1px;
+                    border-radius: 6px;
+                    box-shadow: 0 2px 4px rgba(255, 235, 59, 0.3);
+                    font-weight: bold;
+                    transition: all 0.3s ease;
+                }}
+                .phrase-text {{
+                    padding: 2px 4px;
+                    margin: 1px 0px;
+                    border-radius: 4px;
+                    transition: all 0.3s ease;
+                }}
+                .reading-container {{
+                    max-height: 40vh;
+                    overflow-y: auto;
+                    padding: 20px;
+                    border: 2px solid #e1e1e1;
+                    border-radius: 12px;
+                    background: #fafafa;
+                    margin: 5px 0;
+                    line-height: 1.8;
+                    font-size: 16px;
+                    font-family: Arial, sans-serif;
+                }}
+                .status-bar {{
+                    background: #e3f2fd;
+                    padding: 8px;
+                    border-radius: 8px;
+                    margin: 8px 0;
+                    font-size: 14px;
+                }}
+                .audio-player {{
+                    width: 100%;
+                    margin: 8px 0;
+                    text-align: center;
+                }}
+                .audio-controls {{
+                    display: flex;
+                    justify-content: center;
+                    align-items: center;
+                    gap: 10px;
+                    margin-bottom: 10px;
+                }}
+                </style>
+            </head>
+            <body>
+                <div class="audio-player">
+                    <div class="audio-controls">
+                        <audio id="mainAudio" controls playbackRate={playback_speed} style="min-width: 250px;">
+                            <source src="data:audio/wav;base64,{audio_b64}" type="audio/wav">
+                        </audio>
+                    </div>
+                    <div style="margin-top: 5px; font-size: 14px; color: #666;">
+                        🎵 Seɣbel, tekki ɣef ▶️ afella i tɣuri s urured: <strong>{playback_speed}x</strong>
+                        {"" if playback_speed == 1.0 else " - " + ("ddac ddac" if playback_speed < 1.0 else "aɣiwel")}
+                    </div>
+                </div>
+                
+
+                
+                <div class="reading-container" id="readingContainer">
+                    {full_text}
+                </div>
+
+                <script>
+                const timingInfo = {timing_json};
+                const fullText = `{full_text}`;
+                const playbackSpeed = {playback_speed};
+                let currentHighlightIndex = -1;
+                let phraseElements = [];
+                
+                function initializeHighlighting() {{
+                    // Create phrase elements by wrapping text
+                    let container = document.getElementById('readingContainer');
+                    let workingText = fullText;
+                    
+                    timingInfo.forEach((phrase, index) => {{
+                        const cleanPhrase = phrase.text.trim();
+                        if (workingText.includes(cleanPhrase)) {{
+                            const spanId = 'phrase_' + index;
+                            const spanHtml = '<span id="' + spanId + '" class="phrase-text">' + cleanPhrase + '</span>';
+                            workingText = workingText.replace(cleanPhrase, spanHtml);
+                        }}
+                    }});
+                    
+                    container.innerHTML = workingText;
+                    
+                    // Store references to all phrase elements
+                    timingInfo.forEach((phrase, index) => {{
+                        const element = document.getElementById('phrase_' + index);
+                        if (element) {{
+                            phraseElements.push(element);
+                        }}
+                    }});
+                    
+                    updateDebugInfo("Agzam amegzu yemmed! " + phraseElements.length + " n tefyar s " + playbackSpeed + "x arured");
+                }}
+                
+                function updateDebugInfo(message) {{
+                    const debugEl = document.getElementById('debugInfo');
+                    if (debugEl) debugEl.textContent = message;
+                }}
+                
+                function highlightCurrentPhrase(currentTime) {{
+                    let newIndex = -1;
+                    for (let i = 0; i < timingInfo.length; i++) {{
+                        if (currentTime >= timingInfo[i].start && currentTime < timingInfo[i].end) {{
+                            newIndex = i;
+                            break;
+                        }}
+                    }}
+                    
+                    if (newIndex !== currentHighlightIndex) {{
+                        currentHighlightIndex = newIndex;
+                        updateHighlightDisplay();
+                        if (newIndex >= 0) {{
+                            updateDebugInfo("Akud: " + currentTime.toFixed(2) + "s | Tafyirt: " + (newIndex + 1) + "/" + timingInfo.length + " | Arured: " + playbackSpeed + "x");
+                        }}
+                    }}
+                }}
+                
+                function updateHighlightDisplay() {{
+                    const currentPhraseSpan = document.getElementById('currentPhrase');
+                    
+                    // Remove all highlights
+                    phraseElements.forEach(element => {{
+                        element.className = 'phrase-text';
+                    }});
+                    
+                    // Highlight current phrase
+                    if (currentHighlightIndex >= 0 && currentHighlightIndex < phraseElements.length) {{
+                        const element = phraseElements[currentHighlightIndex];
+                        if (element) {{
+                            element.className = 'phrase-highlight';
+                            element.scrollIntoView({{ behavior: 'smooth', block: 'center' }});
+                        }}
+                        
+                        if (currentPhraseSpan && timingInfo[currentHighlightIndex]) {{
+                            currentPhraseSpan.textContent = timingInfo[currentHighlightIndex].text.substring(0, 100) + 
+                                (timingInfo[currentHighlightIndex].text.length > 100 ? '...' : '');
+                        }}
+                    }} else {{
+                        if (currentPhraseSpan) {{
+                            currentPhraseSpan.textContent = 'Araǧu amesli...';
+                        }}
+                    }}
+                }}
+                
+                // Set up audio event listeners
+                function setupAudioListeners() {{
+                    const audioElement = document.getElementById('mainAudio');
+                    if (audioElement) {{
+                        // Set playback rate
+                        audioElement.playbackRate = playbackSpeed;
+                        
+                        audioElement.addEventListener('timeupdate', function() {{
+                            highlightCurrentPhrase(this.currentTime);
+                        }});
+                        
+                        audioElement.addEventListener('play', function() {{
+                            updateDebugInfo("🎵 Taɣuri... aseḍfeṛ n tira iteddu s " + playbackSpeed + "x arured");
+                        }});
+                        
+                        audioElement.addEventListener('ended', function() {{
+                            currentHighlightIndex = -1;
+                            updateHighlightDisplay();
+                            updateDebugInfo("✅ Taɣuri tekfa s " + playbackSpeed + "x arured");
+                        }});
+                        
+                    }} else {{
+                        setTimeout(setupAudioListeners, 100);
+                    }}
+                }}
+                
+                // Initialize everything when page loads
+                document.addEventListener('DOMContentLoaded', function() {{
+                    initializeHighlighting();
+                    setupAudioListeners();
+                }});
+                </script>
+            </body>
+            </html>
+            """
+            
+            # Display the complete reading content
+            st.components.v1.html(complete_html, height=300, scrolling=True)
+            
+            # Place the remaining controls BELOW the reading content
+            st.markdown('<div class="controls-section">', unsafe_allow_html=True)
+            
+            # Show paragraph info
+            word_count = current_data.get('word_count', len(current_data['paragraph_text'].split()))
+            st.markdown(f"**Taseddaṛt {current_index + 1}/{st.session_state.total_paragraphs}**")
+            st.caption(f"📊 {word_count} n wawalen | ⏱️ {current_data['audio_duration']:.1f}s | 🎵 {st.session_state.playback_speed}x arured")
+
+            # Display progress
+            ready_count = len(st.session_state.paragraphs_data)
+            total_count = st.session_state.total_paragraphs
+            progress = ready_count / total_count if total_count > 0 else 0
+            st.progress(progress)
+            st.caption(f"📊 Asekker: {ready_count}/{total_count} n tseddarin mmedent ({progress:.0%})")
+            
+            # Download button for current paragraph
+            audio_bytes = current_data['audio_bytes']
+            st.download_button(
+                "📥 Zdem ameslaw n tseddaṛt-a",
+                audio_bytes,
+                f"Taseddaṛt_{current_index + 1}.wav",
+                "audio/wav",
+                use_container_width=True
+            )
+            
+            st.markdown('</div>', unsafe_allow_html=True)
+
+        elif uploaded_file and not st.session_state.processed:
+            if st.button("🔄 Selket Aḍris", type="primary"):
+                # Process document when button is clicked
+                with st.spinner("Asekker n uḍris s ugzam amegzu n tseddaṛin..."):
+                    temp_path = os.path.join(tempfile.gettempdir(), uploaded_file.name)
+                    with open(temp_path, "wb") as f:
+                        f.write(uploaded_file.getvalue())
+
+                    text, error = read_document(temp_path)
+                    if error:
+                        st.error(error)
+                    else:
+                        cleaned_text = clean_text(text)
+                        st.session_state.full_document_text = cleaned_text
+                        
+                        # Use smart splitting strategy
+                        paragraphs = smart_split_paragraphs(
+                            cleaned_text, 
+                            initial_paragraphs=initial_paragraphs,
+                            initial_word_target=initial_word_target,
+                            normal_word_target=normal_word_target
+                        )
+                        
+                        if not paragraphs:
+                            st.error("Ulac agbur i tɣuri.")
+                            return
+                        
+                        # Initialize processing state
+                        st.session_state.total_paragraphs = len(paragraphs)
+                        st.session_state.current_paragraph_index = 0
+                        st.session_state.paragraphs_data = {}
+                        st.session_state.paragraphs_list = paragraphs
+                        st.session_state.processed = True
+                        
+                        # Generate first paragraph immediately in main thread
+                        first_paragraph = paragraphs[0]
+                        audio_data, sampling_rate, timing_info = generate_audio_with_precise_timing(first_paragraph)
+                        
+                        st.session_state.paragraphs_data[0] = {
+                            'paragraph_text': first_paragraph,
+                            'audio_data': audio_data,
+                            'timing_info': timing_info,
+                            'audio_duration': len(audio_data) / sampling_rate,
+                            'audio_bytes': get_audio_bytes(audio_data),
+                            'word_count': len(first_paragraph.split())
+                        }
+                        st.session_state.audio_ready = True
+                        
+                        # Start background worker for ALL remaining paragraphs
+                        if len(paragraphs) > 1:
+                            remaining_paragraphs = paragraphs[1:]
+                            
+                            # Use queue-based background worker
+                            thread = threading.Thread(
+                                target=background_audio_worker,
+                                args=(remaining_paragraphs, st.session_state.audio_queue, 1)
+                            )
+                            thread.daemon = True
+                            thread.start()
+                            
+                            st.session_state.background_worker_started = True
+                        
+                        st.rerun()
+            else:
+                st.info("🔄 Seɣbel, tekki ɣef 'Selket Aḍris' iwakken ad yettwasleḍ u ad yeddu seg tira ɣer umeslaw")
+        else:
+            st.info("👆 Sali-d afaylu iwakken ad tedduḍ ɣer tɣuri")
+
+if __name__ == "__main__":
     main()