streamlit_app.py

import streamlit as st
import requests
import tempfile
import os
import subprocess
import speech_recognition as sr
from pydub import AudioSegment
import re
from typing import Dict, Tuple
import time

# Configure Streamlit page
st.set_page_config(
    page_title="English Accent Detector | REM Waste",
    page_icon="🎤",
    layout="wide",
    initial_sidebar_state="collapsed"
)

# Custom CSS for better styling
st.markdown("""
<style>
    .main > div {
        padding-top: 2rem;
    }
    .stButton > button {
        width: 100%;
        border-radius: 10px;
        border: none;
        background: linear-gradient(90deg, #667eea 0%, #764ba2 100%);
        color: white;
        font-weight: bold;
        padding: 0.75rem;
    }
    .metric-container {
        background: #f0f2f6;
        padding: 1rem;
        border-radius: 10px;
        text-align: center;
    }
</style>
""", unsafe_allow_html=True)

class AccentDetector:
    """Streamlined accent detection for English speech analysis"""
    
    def __init__(self):
        self.accent_patterns = {
            'American': {
                'keywords': ['gonna', 'wanna', 'gotta', 'kinda', 'sorta', 'yeah', 'awesome', 'dude'],
                'vocabulary': ['elevator', 'apartment', 'garbage', 'vacation', 'cookie', 'candy', 'mom', 'color'],
                'phrases': ['you know', 'like totally', 'for sure', 'right now']
            },
            'British': {
                'keywords': ['brilliant', 'lovely', 'quite', 'rather', 'chap', 'bloody', 'bloke', 'cheers'],
                'vocabulary': ['lift', 'flat', 'rubbish', 'holiday', 'biscuit', 'queue', 'mum', 'colour'],
                'phrases': ['i say', 'good heavens', 'how do you do', 'spot on']
            },
            'Australian': {
                'keywords': ['mate', 'bloody', 'crikey', 'reckon', 'fair dinkum', 'bonkers', 'ripper'],
                'vocabulary': ['arvo', 'brekkie', 'servo', 'bottle-o', 'mozzie', 'barbie', 'ute'],
                'phrases': ['no worries', 'good on ya', 'she\'ll be right', 'too right']
            },
            'Canadian': {
                'keywords': ['eh', 'about', 'house', 'out', 'sorry', 'hoser', 'beauty'],
                'vocabulary': ['toque', 'hydro', 'washroom', 'parkade', 'chesterfield', 'serviette'],
                'phrases': ['you bet', 'take off', 'give\'r', 'double double']
            },
            'South African': {
                'keywords': ['ag', 'man', 'hey', 'lekker', 'eish', 'shame', 'howzit'],
                'vocabulary': ['robot', 'bakkie', 'boerewors', 'biltong', 'braai', 'veld'],
                'phrases': ['just now', 'now now', 'is it', 'sharp sharp']
            }
        }
    
    @st.cache_data
    def download_video(_self, url: str) -> str:
        """Download video with caching, including Loom/YouTube support and debug output"""
        try:
            headers = {
                'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36'
            }
            # YouTube support (including Shorts)
            if 'youtube.com' in url or 'youtu.be' in url:
                try:
                    import yt_dlp
                except ImportError:
                    raise Exception("yt-dlp is required for YouTube downloads. Please install with 'pip install yt-dlp'.")
                # Use yt-dlp to download best audio to a temp directory, let yt-dlp pick the filename
                tmpdir = tempfile.mkdtemp()
                ydl_opts = {
                    'format': 'bestaudio[ext=m4a]/bestaudio/best',
                    'outtmpl': f'{tmpdir}/%(id)s.%(ext)s',
                    'quiet': True,
                    'noplaylist': True,
                    'postprocessors': [{
                        'key': 'FFmpegExtractAudio',
                        'preferredcodec': 'wav',
                        'preferredquality': '192',
                    }],
                    'ffmpeg_location': '/opt/homebrew/bin/ffmpeg',
                    'overwrites': True,
                }
                try:
                    with yt_dlp.YoutubeDL(ydl_opts) as ydl:
                        info = ydl.extract_info(url, download=True)
                    # Find the resulting .wav file
                    for f in os.listdir(tmpdir):
                        if f.endswith('.wav'):
                            # Move the file to a permanent temp location so it persists after this function
                            final_temp = tempfile.NamedTemporaryFile(delete=False, suffix='.wav')
                            final_temp.close()
                            with open(os.path.join(tmpdir, f), 'rb') as src, open(final_temp.name, 'wb') as dst:
                                dst.write(src.read())
                            return final_temp.name
                    raise Exception("yt-dlp did not produce a valid audio file. Try another video or update yt-dlp/ffmpeg.")
                except Exception as e:
                    raise Exception(f"yt-dlp failed: {str(e)}. Try updating yt-dlp and ffmpeg.")
            # Loom support (fallback: try to extract video from page HTML)
            if 'loom.com' in url:
                resp = requests.get(url, headers=headers, timeout=30)
                if resp.status_code != 200:
                    raise Exception("Failed to fetch Loom page")
                html = resp.text
                import re
                match = re.search(r'src="([^"]+\.mp4)"', html)
                if not match:
                    match = re.search(r'https://cdn\.loom\.com/sessions/[^"\s]+\.mp4', html)
                if not match:
                    raise Exception("Could not extract Loom video stream URL from page HTML")
                video_url = match.group(1)
                url = video_url
            # Download video (Loom or direct)
            response = requests.get(url, headers=headers, stream=True, timeout=60)
            response.raise_for_status()
            with tempfile.NamedTemporaryFile(delete=False, suffix='.mp4') as temp_file:
                for chunk in response.iter_content(chunk_size=8192):
                    if chunk:
                        temp_file.write(chunk)
                return temp_file.name
        except Exception as e:
            raise Exception(f"Download failed: {str(e)}")
    
    def extract_audio_simple(self, video_path: str) -> str:
        """Robust audio extraction: handles mp3, wav, mp4, etc."""
        try:
            import os
            from pydub import AudioSegment
            ext = os.path.splitext(video_path)[1].lower()
            audio_path = video_path.rsplit('.', 1)[0] + '.wav'
            # If already wav, use pydub directly
            if ext == '.wav':
                audio = AudioSegment.from_wav(video_path)
            else:
                audio = AudioSegment.from_file(video_path)
            audio = audio.set_frame_rate(16000).set_channels(1)
            if len(audio) > 120000:
                audio = audio[:120000]
            audio.export(audio_path, format="wav")
            return audio_path
        except Exception as e:
            raise Exception(f"Audio extraction failed: {str(e)}")
    
    def transcribe_audio(self, audio_path: str) -> str:
        """Transcribe with error handling"""
        try:
            r = sr.Recognizer()
            r.energy_threshold = 300
            r.dynamic_energy_threshold = True
            
            with sr.AudioFile(audio_path) as source:
                r.adjust_for_ambient_noise(source, duration=0.5)
                audio_data = r.record(source)
            
            # Try Google Speech Recognition
            text = r.recognize_google(audio_data, language='en-US')
            return text.lower()
            
        except sr.UnknownValueError:
            raise Exception("Could not understand the audio clearly")
        except sr.RequestError as e:
            raise Exception(f"Speech recognition service error: {str(e)}")
        except Exception as e:
            raise Exception(f"Transcription failed: {str(e)}")
    
    def analyze_patterns(self, text: str) -> Dict[str, float]:
        """Enhanced pattern analysis"""
        scores = {}
        words = text.split()
        word_count = max(len(words), 1)
        
        for accent, patterns in self.accent_patterns.items():
            score = 0.0
            total_matches = 0
            
            # Keywords (high weight)
            for keyword in patterns['keywords']:
                if keyword in text:
                    score += 20.0
                    total_matches += 1
            
            # Vocabulary (medium weight)  
            for vocab in patterns['vocabulary']:
                if vocab in text:
                    score += 15.0
                    total_matches += 1
            
            # Phrases (high weight)
            for phrase in patterns['phrases']:
                if phrase in text:
                    score += 25.0
                    total_matches += 1
            
            # Normalize and add base confidence
            if total_matches > 0:
                score = min(score * (total_matches / word_count) * 50, 95.0)
            else:
                score = self._get_base_score(text, accent)
            
            scores[accent] = round(max(score, 5.0), 1)
        
        return scores
    
    def _get_base_score(self, text: str, accent: str) -> float:
        """Base scoring for general patterns"""
        base_scores = {
            'American': 30.0,
            'British': 20.0, 
            'Australian': 15.0,
            'Canadian': 18.0,
            'South African': 12.0
        }
        
        score = base_scores.get(accent, 15.0)
        
        # Spelling adjustments
        if accent == 'British':
            if any(word in text for word in ['colour', 'favour', 'centre', 'theatre']):
                score += 25.0
        elif accent == 'American':
            if any(word in text for word in ['color', 'favor', 'center', 'theater']):
                score += 25.0
        
        return min(score, 45.0)
    
    def classify_accent(self, scores: Dict[str, float]) -> Tuple[str, float, str]:
        """Classify and explain results"""
        if not scores:
            return "Unknown", 0.0, "No speech detected"
        
        # Get top result
        top_accent = max(scores.items(), key=lambda x: x[1])
        accent, confidence = top_accent
        
        # Generate explanation
        if confidence < 25:
            explanation = "Low confidence - speech patterns are not strongly distinctive"
        elif confidence < 50:
            explanation = f"Moderate confidence in {accent} accent based on some linguistic markers"
        elif confidence < 75:
            explanation = f"Good confidence in {accent} accent with clear characteristic patterns"
        else:
            explanation = f"High confidence in {accent} accent with strong linguistic evidence"
        
        return accent, confidence, explanation

# Initialize detector
@st.cache_resource
def get_detector():
    return AccentDetector()

def main():
    # Header
    st.title("🎤 English Accent Detection Tool")
    st.markdown("**AI-powered accent analysis for English speech | Built for REM Waste**")
    
    # Description
    with st.expander("ℹ️ How it works", expanded=False):
        st.markdown("""
        1. **Input**: Paste a public video URL (MP4, Loom, YouTube, etc.)
        2. **Processing**: Extract audio → Transcribe speech → Analyze patterns
        3. **Output**: Accent classification + confidence score + explanation
        
        **Supported Accents**: American, British, Australian, Canadian, South African
        """)
    
    # Input section
    st.subheader("📹 Video Input")
    
    # Sample URLs for testing
    with st.expander("🧪 Test with sample videos"):
        st.markdown("""
        **Sample URLs for testing:**
        - `https://sample-videos.com/zip/10/mp4/SampleVideo_1280x720_1mb.mp4`
        - `https://www.learningcontainer.com/wp-content/uploads/2020/05/sample-mp4-file.mp4`
        - Or any public Loom/YouTube video URL
        """)
    
    video_url = st.text_input(
        "Enter video URL:",
        placeholder="https://example.com/video.mp4",
        help="Must be a publicly accessible video URL"
    )
    
    # Process button
    if st.button("🚀 Analyze Accent", type="primary"):
        if not video_url.strip():
            st.error("⚠️ Please enter a video URL")
            return
        
        if not video_url.startswith(('http://', 'https://')):
            st.error("⚠️ Please enter a valid URL starting with http:// or https://")
            return
        
        # Initialize detector and progress tracking
        detector = get_detector()
        temp_files = []
        
        try:
            # Progress bar
            progress_bar = st.progress(0)
            status_text = st.empty()
            
            # Step 1: Download video
            status_text.text("📥 Downloading video...")
            progress_bar.progress(20)
            video_path = detector.download_video(video_url)
            temp_files.append(video_path)
            
            # Step 2: Extract audio
            status_text.text("🎵 Extracting audio...")
            progress_bar.progress(50)
            audio_path = detector.extract_audio_simple(video_path)
            temp_files.append(audio_path)
            
            # Step 3: Transcribe
            status_text.text("🎤 Transcribing speech...")
            progress_bar.progress(75)
            transcript = detector.transcribe_audio(audio_path)
            
            # Step 4: Analyze
            status_text.text("🔍 Analyzing accent patterns...")
            progress_bar.progress(90)
            scores = detector.analyze_patterns(transcript)
            accent, confidence, explanation = detector.classify_accent(scores)
            
            # Complete
            progress_bar.progress(100)
            status_text.text("✅ Analysis complete!")
            time.sleep(0.5)
            
            # Clear progress indicators
            progress_bar.empty()
            status_text.empty()
            
            # Display results
            st.success("🎉 **Analysis Complete!**")
            
            # Main metrics
            col1, col2, col3 = st.columns(3)
            
            with col1:
                st.markdown(f"""
                <div class="metric-container">
                    <h3>🗣️ Detected Accent</h3>
                    <h2 style="color: #667eea;">{accent}</h2>
                </div>
                """, unsafe_allow_html=True)
            
            with col2:
                st.markdown(f"""
                <div class="metric-container">
                    <h3>🎯 Confidence</h3>
                    <h2 style="color: #764ba2;">{confidence}%</h2>
                </div>
                """, unsafe_allow_html=True)
            
            with col3:
                # Get transcript length for quality indicator
                word_count = len(transcript.split())
                quality = "High" if word_count > 50 else "Medium" if word_count > 20 else "Low"
                st.markdown(f"""
                <div class="metric-container">
                    <h3>📊 Data Quality</h3>
                    <h2 style="color: #28a745;">{quality}</h2>
                    <small>{word_count} words</small>
                </div>
                """, unsafe_allow_html=True)
            
            st.markdown("---")
            
            # Explanation
            st.subheader("📝 Analysis Summary")
            st.info(explanation)
            
            # Transcript
            st.subheader("📄 Transcribed Speech")
            st.text_area(
                "Full transcript:",
                transcript,
                height=120,
                help="This is what the AI heard from the video"
            )
            
            # Detailed scores
            st.subheader("📊 All Accent Scores")
            
            # Create a more visual representation
            for accent_name, score in sorted(scores.items(), key=lambda x: x[1], reverse=True):
                # Create progress bar for each accent
                col_name, col_bar, col_score = st.columns([2, 6, 1])
                
                with col_name:
                    st.write(f"**{accent_name}**")
                
                with col_bar:
                    st.progress(score / 100)
                
                with col_score:
                    st.write(f"{score}%")
            
            # Additional insights
            if confidence > 60:
                st.success(f"🎯 **Strong Detection**: The {accent} accent markers are clearly present in the speech.")
            elif confidence > 40:
                st.warning(f"⚠️ **Moderate Detection**: Some {accent} patterns detected, but results may vary with longer audio.")
            else:
                st.info("💡 **Tip**: Longer speech samples (30+ seconds) generally provide more accurate results.")
            
        except Exception as e:
            st.error(f"❌ **Processing Error**: {str(e)}")
            st.info("""
            **Troubleshooting Tips:**
            - Ensure the video URL is publicly accessible
            - Try a different video format or shorter video
            - Make sure the video contains clear English speech
            - Check your internet connection
            """)
        
        finally:
            # Cleanup temp files
            for temp_file in temp_files:
                try:
                    if os.path.exists(temp_file):
                        os.remove(temp_file)
                except:
                    pass
    
    # Footer information
    st.markdown("---")
    
    col1, col2 = st.columns(2)
    
    with col1:
        st.markdown("""
        **🔧 Technical Details**
        - Audio processing: Up to 2 minutes
        - Speech recognition: Google API
        - Analysis: Pattern matching + linguistics
        - Processing time: ~30-90 seconds
        """)
    
    with col2:
        st.markdown("""
        **📋 Requirements**
        - Public video URLs only
        - Clear English speech preferred
        - Supports MP4, MOV, AVI formats
        - Works with Loom, YouTube, direct links
        """)
    
    # API information
    with st.expander("🔗 API Usage"):
        st.code("""
# Python API usage example
from accent_detector import AccentDetector

detector = AccentDetector()
result = detector.process_video("https://your-video.com/file.mp4")

print(f"Accent: {result['accent']}")
print(f"Confidence: {result['confidence']}%")
        """, language="python")
    
    # About section
    with st.expander("ℹ️ About This Tool"):
        st.markdown("""
        **Built for REM Waste Interview Challenge**
        
        This accent detection tool analyzes English speech patterns to classify regional accents. 
        It's designed for hiring automation systems that need to evaluate spoken English proficiency.
        
        **Algorithm Overview:**
        - Extracts audio from video files
        - Transcribes speech using Google Speech Recognition
        - Analyzes linguistic patterns, vocabulary, and pronunciation markers
        - Provides confidence scores based on pattern strength
        
        **Accuracy Notes:**
        - Best results with 30+ seconds of clear speech
        - Confidence scores reflect pattern strength, not absolute accuracy
        - Designed for screening purposes, not definitive classification
        
        **Privacy & Ethics:**
        - No audio/video data is stored permanently
        - Temporary files are automatically deleted
        - Tool is intended for voluntary language assessment only
        """)

if __name__ == "__main__":
    main()