Initial commit with clean history (no audio files)

.gitattributes

@@ -0,0 +1,36 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+*.wav filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,38 @@
+# Python bytecode
+__pycache__/
+*.py[cod]
+*$py.class
+
+# Virtual environment
+venv/
+env/
+ENV/
+
+# Distribution / packaging
+dist/
+build/
+*.egg-info/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# Temp directory
+/tmp/
+
+# Log files
+*.log
+
+# Mac OS files
+.DS_Store
+
+# Temp audio files
+*.wav.tmp
+
+# Audio files
+*.wav
+
+# Exclude data directories except README files
+torgo_original/data/*
+!torgo_original/data/README.md
+torgo-synthetic/data/*
+!torgo-synthetic/data/README.md 

.gradio/certificate.pem

@@ -0,0 +1,31 @@
+-----BEGIN CERTIFICATE-----
+MIIFazCCA1OgAwIBAgIRAIIQz7DSQONZRGPgu2OCiwAwDQYJKoZIhvcNAQELBQAw
+TzELMAkGA1UEBhMCVVMxKTAnBgNVBAoTIEludGVybmV0IFNlY3VyaXR5IFJlc2Vh
+cmNoIEdyb3VwMRUwEwYDVQQDEwxJU1JHIFJvb3QgWDEwHhcNMTUwNjA0MTEwNDM4
+WhcNMzUwNjA0MTEwNDM4WjBPMQswCQYDVQQGEwJVUzEpMCcGA1UEChMgSW50ZXJu
+ZXQgU2VjdXJpdHkgUmVzZWFyY2ggR3JvdXAxFTATBgNVBAMTDElTUkcgUm9vdCBY
+MTCCAiIwDQYJKoZIhvcNAQEBBQADggIPADCCAgoCggIBAK3oJHP0FDfzm54rVygc
+h77ct984kIxuPOZXoHj3dcKi/vVqbvYATyjb3miGbESTtrFj/RQSa78f0uoxmyF+
+0TM8ukj13Xnfs7j/EvEhmkvBioZxaUpmZmyPfjxwv60pIgbz5MDmgK7iS4+3mX6U
+A5/TR5d8mUgjU+g4rk8Kb4Mu0UlXjIB0ttov0DiNewNwIRt18jA8+o+u3dpjq+sW
+T8KOEUt+zwvo/7V3LvSye0rgTBIlDHCNAymg4VMk7BPZ7hm/ELNKjD+Jo2FR3qyH
+B5T0Y3HsLuJvW5iB4YlcNHlsdu87kGJ55tukmi8mxdAQ4Q7e2RCOFvu396j3x+UC
+B5iPNgiV5+I3lg02dZ77DnKxHZu8A/lJBdiB3QW0KtZB6awBdpUKD9jf1b0SHzUv
+KBds0pjBqAlkd25HN7rOrFleaJ1/ctaJxQZBKT5ZPt0m9STJEadao0xAH0ahmbWn
+OlFuhjuefXKnEgV4We0+UXgVCwOPjdAvBbI+e0ocS3MFEvzG6uBQE3xDk3SzynTn
+jh8BCNAw1FtxNrQHusEwMFxIt4I7mKZ9YIqioymCzLq9gwQbooMDQaHWBfEbwrbw
+qHyGO0aoSCqI3Haadr8faqU9GY/rOPNk3sgrDQoo//fb4hVC1CLQJ13hef4Y53CI
+rU7m2Ys6xt0nUW7/vGT1M0NPAgMBAAGjQjBAMA4GA1UdDwEB/wQEAwIBBjAPBgNV
+HRMBAf8EBTADAQH/MB0GA1UdDgQWBBR5tFnme7bl5AFzgAiIyBpY9umbbjANBgkq
+hkiG9w0BAQsFAAOCAgEAVR9YqbyyqFDQDLHYGmkgJykIrGF1XIpu+ILlaS/V9lZL
+ubhzEFnTIZd+50xx+7LSYK05qAvqFyFWhfFQDlnrzuBZ6brJFe+GnY+EgPbk6ZGQ
+3BebYhtF8GaV0nxvwuo77x/Py9auJ/GpsMiu/X1+mvoiBOv/2X/qkSsisRcOj/KK
+NFtY2PwByVS5uCbMiogziUwthDyC3+6WVwW6LLv3xLfHTjuCvjHIInNzktHCgKQ5
+ORAzI4JMPJ+GslWYHb4phowim57iaztXOoJwTdwJx4nLCgdNbOhdjsnvzqvHu7Ur
+TkXWStAmzOVyyghqpZXjFaH3pO3JLF+l+/+sKAIuvtd7u+Nxe5AW0wdeRlN8NwdC
+jNPElpzVmbUq4JUagEiuTDkHzsxHpFKVK7q4+63SM1N95R1NbdWhscdCb+ZAJzVc
+oyi3B43njTOQ5yOf+1CceWxG1bQVs5ZufpsMljq4Ui0/1lvh+wjChP4kqKOJ2qxq
+4RgqsahDYVvTH9w7jXbyLeiNdd8XM2w9U/t7y0Ff/9yi0GE44Za4rF2LN9d11TPA
+mRGunUHBcnWEvgJBQl9nJEiU0Zsnvgc/ubhPgXRR4Xq37Z0j4r7g1SgEEzwxA57d
+emyPxgcYxn/eR44/KJ4EBs+lVDR3veyJm+kXQ99b21/+jh5Xos1AnX5iItreGCc=
+-----END CERTIFICATE-----

README.md

@@ -0,0 +1,39 @@
+---
+title: Speech Evaluation Experiment
+emoji: 👀
+colorFrom: yellow
+colorTo: pink
+sdk: gradio
+sdk_version: 5.31.0
+app_file: app.py
+pinned: false
+license: apache-2.0
+---
+
+# Speech Evaluation Experiment
+
+This application allows users to evaluate synthetic speech samples against original samples, rating their naturalness and intelligibility.
+
+## Requirements
+
+All dependencies are listed in `requirements.txt`.
+
+## Setup
+
+1. Clone this repository
+2. Install dependencies: `pip install -r requirements.txt`
+3. Run the application: `python experiment1.py`
+
+## Data Structure
+
+The experiment uses two datasets:
+- `torgo_original` - Contains original speech samples
+- `torgo-synthetic` - Contains synthetic speech samples
+
+Each dataset has its own metadata.csv file that describes the audio files.
+
+## Deployment
+
+This application is designed to work with Hugging Face Spaces or similar platforms that support Gradio applications.
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,12 @@
+import os
+from experiment1 import create_experiment_interface
+
+# Create the Gradio interface
+demo = create_experiment_interface()
+
+# Launch the app
+if __name__ == "__main__":
+    # For local development
+    demo.launch()
+    
+# For Hugging Face Spaces, the demo variable will be used automatically 

deploy_instructions.md

@@ -0,0 +1,46 @@
+# Deployment Instructions for Hugging Face Spaces
+
+Follow these steps to deploy your speech evaluation app to Hugging Face Spaces:
+
+## 1. Create a Hugging Face Account
+- Go to https://huggingface.co/join if you don't have an account
+
+## 2. Create a New Space
+- Go to https://huggingface.co/spaces
+- Click "Create new Space"
+- Choose "Gradio" as the SDK
+- Name your Space (e.g., "speech-evaluation-experiment")
+- Set visibility (Public or Private)
+- Click "Create Space"
+
+## 3. Push Your Code to the Space
+Run these commands in your terminal:
+
+```bash
+# Initialize git repository (if not already done)
+git init
+
+# Add all files
+git add .
+
+# Commit changes
+git commit -m "Initial commit for deployment"
+
+# Add Hugging Face Space as remote (replace YOUR_USERNAME with your Hugging Face username)
+git remote add space https://huggingface.co/spaces/YOUR_USERNAME/speech-evaluation-experiment
+
+# Push to Hugging Face
+git push --force space main
+```
+
+## 4. Monitor Deployment
+- Go to your Space at https://huggingface.co/spaces/YOUR_USERNAME/speech-evaluation-experiment
+- You can see build logs and when your app is deployed
+
+## 5. Custom Domain (Optional)
+- You can set up a custom domain in the Space settings
+
+## Important Notes
+- Make sure all audio files are included in the repository
+- The app.py file will be automatically detected and run
+- Requirements.txt will be used to install dependencies 

deploy_to_huggingface.md

@@ -0,0 +1,42 @@
+# Deploy to Hugging Face Spaces with Token Authentication
+
+## Step 1: Create a Hugging Face Access Token
+
+1. Go to https://huggingface.co/settings/tokens
+2. Click "New token"
+3. Give it a name (e.g., "Speech Evaluation App")
+4. Set permissions to "Write"
+5. Click "Generate a token"
+6. Copy the token (you'll only see it once)
+
+## Step 2: Deploy to Hugging Face Spaces
+
+Run these commands in your terminal:
+
+```bash
+# Initialize git repository
+git init
+
+# Add all files
+git add .
+
+# Commit changes
+git commit -m "Initial deployment"
+
+# Add Hugging Face Space as remote
+git remote add origin https://huggingface.co/spaces/kesbeast23/speech-evaluation-experiment
+
+# Push to Hugging Face
+git push -u origin main
+```
+
+When prompted for username and password:
+- Username: your Hugging Face username
+- Password: paste your access token (not your account password)
+
+## Step 3: Check Deployment
+
+Once the push is complete, visit your Space at:
+https://huggingface.co/spaces/kesbeast23/speech-evaluation-experiment
+
+It may take a few minutes for the app to build and deploy. 

experiment1.py

@@ -0,0 +1,662 @@
+import gradio as gr
+import os
+import random
+import pandas as pd
+from datetime import datetime
+import numpy as np
+import uuid
+import soundfile as sf
+import librosa
+import noisereduce as nr
+import tempfile
+import atexit
+import shutil
+
+# Constants
+# Get absolute paths
+WORKSPACE_ROOT = os.path.dirname(os.path.abspath(__file__))
+ORIGINAL_DATA_DIR = os.path.join(WORKSPACE_ROOT, "torgo_original")
+SYNTHETIC_DATA_DIR = os.path.join(WORKSPACE_ROOT, "torgo-synthetic")
+RESULTS_FILE = os.path.join(WORKSPACE_ROOT, "experiment_Results.csv")
+TEMP_DIR = os.path.join(tempfile.gettempdir(), "speech_evaluation")
+
+# Create directories if they don't exist
+os.makedirs(TEMP_DIR, exist_ok=True)
+os.makedirs(os.path.join(ORIGINAL_DATA_DIR, "data"), exist_ok=True)
+os.makedirs(os.path.join(SYNTHETIC_DATA_DIR, "data"), exist_ok=True)
+
+# Track generated temp files for cleanup
+temp_files = []
+
+# Flag to check if running in demo mode (no audio files)
+DEMO_MODE = True
+
+# Check if we're in demo mode (no audio files)
+def check_demo_mode():
+    original_data_path = os.path.join(ORIGINAL_DATA_DIR, "data")
+    synthetic_data_path = os.path.join(SYNTHETIC_DATA_DIR, "data")
+    
+    # Check if data directories exist and contain files
+    if (os.path.exists(original_data_path) and len(os.listdir(original_data_path)) > 0 and
+        os.path.exists(synthetic_data_path) and len(os.listdir(synthetic_data_path)) > 0):
+        return False
+    return True
+
+# Set demo mode flag
+DEMO_MODE = check_demo_mode()
+if DEMO_MODE:
+    print("Running in DEMO MODE - No audio files found")
+
+# Register cleanup function to run on exit
+def cleanup_temp_files():
+    """Remove temporary files and directory on exit"""
+    for temp_file in temp_files:
+        try:
+            if os.path.exists(temp_file):
+                os.remove(temp_file)
+        except Exception as e:
+            print(f"Error removing temp file {temp_file}: {e}")
+    
+    try:
+        if os.path.exists(TEMP_DIR):
+            shutil.rmtree(TEMP_DIR)
+    except Exception as e:
+        print(f"Error removing temp directory {TEMP_DIR}: {e}")
+
+atexit.register(cleanup_temp_files)
+
+# Sample type mapping
+SAMPLE_TYPE_MAPPING = {
+    "Original": "Natural",  # For display purposes
+    "Natural": "Original"   # For database storage
+}
+
+# Define columns for results DataFrame
+COLUMNS = [
+    'timestamp', 'participant_id', 'sample_id', 'sample_type',
+    'naturalness_rating', 'intelligibility_rating', 'comments',
+    'transcription', 'original_speaker', 'synthetic_speaker',
+    'participant_guess', 'guess_correct'
+]
+
+# Initialize results DataFrame
+try:
+    results_df = pd.read_csv(RESULTS_FILE)
+    # Verify columns match expected structure
+    if list(results_df.columns) != COLUMNS:
+        results_df = pd.DataFrame(columns=COLUMNS)
+        results_df.to_csv(RESULTS_FILE, index=False)
+except (pd.errors.EmptyDataError, FileNotFoundError):
+    # Create new DataFrame if file is empty or doesn't exist
+    results_df = pd.DataFrame(columns=COLUMNS)
+    results_df.to_csv(RESULTS_FILE, index=False)
+
+# Read metadata files
+original_metadata = pd.read_csv(os.path.join(ORIGINAL_DATA_DIR, "metadata.csv"))
+synthetic_metadata = pd.read_csv(os.path.join(SYNTHETIC_DATA_DIR, "metadata.csv"))
+
+# Set a fixed random seed for reproducibility
+RANDOM_SEED = 42
+random.seed(RANDOM_SEED)
+
+def convert_display_type_to_storage(display_type):
+    """Convert display sample type to storage type"""
+    if display_type == "Natural":
+        return "Original"
+    return display_type
+
+def convert_storage_type_to_display(storage_type):
+    """Convert storage sample type to display type"""
+    if storage_type == "Original":
+        return "Natural"
+    return storage_type
+
+def get_audio_path(file_path, is_original=True):
+    """Convert metadata file path to actual audio file path"""
+    # Remove 'data/' prefix if present
+    file_path = file_path.replace('data/', '')
+    
+    # Construct absolute path
+    if is_original:
+        return os.path.join(ORIGINAL_DATA_DIR, "data", file_path)
+    else:
+        return os.path.join(SYNTHETIC_DATA_DIR, "data", file_path)
+
+def verify_audio_file(file_path):
+    """Verify that audio file exists and is readable"""
+    if DEMO_MODE:
+        # In demo mode, pretend all files exist
+        return True
+        
+    try:
+        if os.path.exists(file_path):
+            data, samplerate = sf.read(file_path)
+            return True
+        return False
+    except:
+        return False
+
+def generate_participant_id():
+    """Generate a unique participant ID"""
+    # Get existing participant IDs
+    existing_ids = set()
+    if os.path.exists(RESULTS_FILE):
+        try:
+            results = pd.read_csv(RESULTS_FILE)
+            if not results.empty:
+                existing_ids = set(results['participant_id'].unique())
+        except pd.errors.EmptyDataError:
+            pass
+    
+    # Find the next available number
+    counter = 1
+    while f"P{counter:03d}" in existing_ids:
+        counter += 1
+    
+    return f"P{counter:03d}"
+
+def preprocess_audio(file_path):
+    """Remove background noise from the audio file and return a temporary file path"""
+    global temp_files
+    
+    if DEMO_MODE:
+        # In demo mode, return a placeholder empty audio file
+        temp_path = os.path.join(TEMP_DIR, f"demo_{uuid.uuid4()}.wav")
+        # Create a short silent wav file
+        sr = 16000
+        silent_audio = np.zeros(int(sr * 1.5))  # 1.5 seconds of silence
+        sf.write(temp_path, silent_audio, sr)
+        temp_files.append(temp_path)
+        return temp_path
+    
+    try:
+        # Load audio file
+        audio, sr = librosa.load(file_path, sr=None)
+        
+        # Apply noise reduction
+        reduced_noise = nr.reduce_noise(y=audio, sr=sr)
+        
+        # Create a temporary file to store the noise-reduced audio
+        temp_path = os.path.join(TEMP_DIR, f"processed_{os.path.basename(file_path)}")
+        sf.write(temp_path, reduced_noise, sr)
+        
+        # Track the temp file for later cleanup
+        temp_files.append(temp_path)
+        
+        return temp_path
+    except Exception as e:
+        print(f"Error preprocessing audio: {e}")
+        # Create a silent audio in case of error
+        temp_path = os.path.join(TEMP_DIR, f"error_{uuid.uuid4()}.wav")
+        sr = 16000
+        silent_audio = np.zeros(int(sr * 1.5))  # 1.5 seconds of silence
+        sf.write(temp_path, silent_audio, sr)
+        temp_files.append(temp_path)
+        return temp_path
+
+# Create a fixed set of original samples and their synthetic versions
+def create_sample_pairs():
+    """Create a selection of original samples and their synthetic versions
+    - Each original sample is included only once
+    - All synthetic versions of each original sample are included"""
+    
+    # Extract speaker IDs from file names (format: data/X_YYY_Session...)
+    original_metadata['speaker_id'] = original_metadata['file_name'].apply(
+        lambda x: x.split('_')[1] if '_' in x else 'unknown'
+    )
+    
+    # Get unique speakers from the extracted speaker IDs
+    original_speakers = original_metadata['speaker_id'].unique()
+    print(f"Found {len(original_speakers)} unique original speakers: {original_speakers}")
+    
+    # First, identify files that have synthetic versions
+    original_files_with_synthetic = synthetic_metadata['original_file'].unique()
+    print(f"Found {len(original_files_with_synthetic)} original files that have synthetic versions")
+    
+    # Group by original file to structure the experiment properly
+    organized_samples = []
+    
+    # Dictionary to store selected original files (to avoid duplicates)
+    selected_original_files = set()
+    
+    # First approach: Select specific example samples from the user's data
+    example_samples = [
+        "data/F_F03_Session3_0164.wav",  # "sing"
+        "data/F_F03_Session3_0170.wav",  # "leak"
+        "data/F_F03_Session3_0158.wav"   # "brought"
+    ]
+    
+    for orig_file in example_samples:
+        if orig_file in original_metadata['file_name'].values and orig_file not in selected_original_files:
+            # Find the transcription for this original file
+            orig_row = original_metadata[original_metadata['file_name'] == orig_file].iloc[0]
+            
+            # Find all synthetic versions of this original file
+            matching_synthetic = synthetic_metadata[
+                synthetic_metadata['original_file'] == orig_file
+            ]
+            
+            if not matching_synthetic.empty:
+                # Verify original file exists
+                orig_path = get_audio_path(orig_file, is_original=True)
+                if verify_audio_file(orig_path):
+                    # Create a group with one original and all its synthetic versions
+                    group = {
+                        'original': {
+                            'file': orig_file,
+                            'path': orig_path,
+                            'transcription': orig_row['transcription'],
+                        },
+                        'synthetic': []
+                    }
+                    
+                    # Add synthetic versions
+                    for _, synth_row in matching_synthetic.iterrows():
+                        synth_path = get_audio_path(synth_row['file_name'], is_original=False)
+                        if verify_audio_file(synth_path):
+                            group['synthetic'].append({
+                                'file': synth_row['file_name'],
+                                'path': synth_path,
+                                'transcription': orig_row['transcription'],
+                                'original_speaker': synth_row['original_speaker'],
+                                'synthetic_speaker': synth_row['synthetic_speaker']
+                            })
+                    
+                    # Only add group if it has synthetic versions
+                    if group['synthetic']:
+                        organized_samples.append(group)
+                        selected_original_files.add(orig_file)
+                        print(f"Added example group: {orig_file} with {len(group['synthetic'])} synthetic versions")
+    
+    # Second approach: If needed, add more samples from other original speakers
+    if len(organized_samples) < 7:  # Aim for at least 7 original samples
+        # Filter original metadata to only include files that have synthetic versions
+        filterable_originals = original_metadata[
+            original_metadata['file_name'].isin(original_files_with_synthetic) & 
+            ~original_metadata['file_name'].isin(selected_original_files)
+        ]
+        
+        # Select samples from each speaker
+        for speaker in original_speakers:
+            # Skip if we already have enough samples
+            if len(organized_samples) >= 7:
+                break
+                
+            speaker_samples = filterable_originals[filterable_originals['speaker_id'] == speaker]
+            
+            # Skip if no samples for this speaker have synthetic versions
+            if len(speaker_samples) == 0:
+                print(f"No additional samples with synthetic versions for speaker {speaker}")
+                continue
+            
+            # Select one sample per speaker
+            selected_sample = speaker_samples.sample(n=1, random_state=RANDOM_SEED).iloc[0]
+            orig_file = selected_sample['file_name']
+            
+            # Skip if already selected
+            if orig_file in selected_original_files:
+                continue
+            
+            # Find all synthetic versions of this original file
+            matching_synthetic = synthetic_metadata[
+                synthetic_metadata['original_file'] == orig_file
+            ]
+            
+            # Verify original file exists
+            orig_path = get_audio_path(orig_file, is_original=True)
+            if verify_audio_file(orig_path):
+                # Create a group with one original and all its synthetic versions
+                group = {
+                    'original': {
+                        'file': orig_file,
+                        'path': orig_path,
+                        'transcription': selected_sample['transcription'],
+                    },
+                    'synthetic': []
+                }
+                
+                # Add synthetic versions
+                for _, synth_row in matching_synthetic.iterrows():
+                    synth_path = get_audio_path(synth_row['file_name'], is_original=False)
+                    if verify_audio_file(synth_path):
+                        group['synthetic'].append({
+                            'file': synth_row['file_name'],
+                            'path': synth_path,
+                            'transcription': selected_sample['transcription'],
+                            'original_speaker': synth_row['original_speaker'],
+                            'synthetic_speaker': synth_row['synthetic_speaker']
+                        })
+                
+                # Only add group if it has synthetic versions
+                if group['synthetic']:
+                    organized_samples.append(group)
+                    selected_original_files.add(orig_file)
+                    print(f"Added additional group: {orig_file} with {len(group['synthetic'])} synthetic versions")
+    
+    # Now flatten the organized samples into a list of samples to play in sequence
+    playback_sequence = []
+    
+    for group in organized_samples:
+        # First add the original
+        playback_sequence.append({
+            'is_original': True,
+            'file_name': group['original']['file'],
+            'file_path': group['original']['path'],
+            'transcription': group['original']['transcription'],
+            'original_speaker': '',
+            'synthetic_speaker': ''
+        })
+        
+        # Then add all synthetic versions
+        for synth in group['synthetic']:
+            playback_sequence.append({
+                'is_original': False,
+                'file_name': synth['file'],
+                'file_path': synth['path'],
+                'transcription': synth['transcription'],
+                'original_speaker': synth['original_speaker'],
+                'synthetic_speaker': synth['synthetic_speaker']
+            })
+    
+    # Print statistics
+    print(f"Created sequence with {len(playback_sequence)} samples:")
+    print(f"- {len(organized_samples)} original samples")
+    print(f"- {len(playback_sequence) - len(organized_samples)} synthetic versions")
+    
+    if len(playback_sequence) == 0:
+        print("WARNING: No samples were created. Please check metadata files.")
+    
+    return playback_sequence
+
+# Initialize sample playback sequence
+playback_sequence = create_sample_pairs()
+print(f"Created playback sequence with {len(playback_sequence)} samples")
+current_sample_index = 0
+
+def save_rating(participant_id, sample_id, sample_type, naturalness, intelligibility, comments, 
+               transcription, original_speaker, synthetic_speaker, participant_guess):
+    """Save the rating to the CSV file"""
+    global results_df  # Move global declaration to the start of the function
+    
+    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+    
+    # Map Yes/No to Natural/Synthetic
+    if participant_guess == "Yes":
+        storage_guess = "Original"
+    elif participant_guess == "No":
+        storage_guess = "Synthetic"
+    else:
+        storage_guess = ""
+    
+    # Check if guess was correct
+    guess_correct = storage_guess.lower() == sample_type.lower()
+    
+    # Get current accuracy
+    participant_results = results_df[results_df['participant_id'] == participant_id]
+    current_accuracy = participant_results['guess_correct'].mean() * 100 if not participant_results.empty else 100
+    
+    # Create feedback message
+    display_type = convert_storage_type_to_display(sample_type)
+    feedback = f"System Feedback: Your guess was {'correct' if guess_correct else 'incorrect'} (It was {display_type}). Current accuracy: {current_accuracy:.1f}%"
+    
+    # Don't combine user comments with feedback - save user comments as is
+    new_row = {
+        'timestamp': timestamp,
+        'participant_id': participant_id,
+        'sample_id': sample_id,
+        'sample_type': sample_type,
+        'naturalness_rating': naturalness,
+        'intelligibility_rating': intelligibility,
+        'comments': comments or "",
+        'transcription': transcription,
+        'original_speaker': original_speaker,
+        'synthetic_speaker': synthetic_speaker,
+        'participant_guess': storage_guess,
+        'guess_correct': guess_correct
+    }
+    
+    results_df = pd.concat([results_df, pd.DataFrame([new_row])], ignore_index=True)
+    results_df.to_csv(RESULTS_FILE, index=False)
+    return feedback
+
+def create_experiment_interface():
+    """Create the Gradio interface for the experiment"""
+    
+    with gr.Blocks(title="Dysarthric Speech Evaluation") as demo:
+        gr.Markdown("""
+        # Dysarthric Speech Evaluation Experiment
+        
+        Welcome to the experiment! You will be asked to evaluate speech samples and determine if they are natural recordings or synthetic (computer-generated) speech.
+        
+        ## Instructions:
+        1. Listen to each audio sample carefully (background noise has been reduced for better clarity)
+        2. Guess whether the sample is Natural (real human recording) or Synthetic (computer-generated)
+        3. Rate the naturalness and intelligibility on a scale of 1-5
+        4. Add any comments about the speech sample (optional)
+        5. Click 'Submit Rating' to save your evaluation and see if your guess was correct
+        
+        ## Rating Scale:
+        - 1: Poor/Unintelligible
+        - 2: Fair
+        - 3: Good
+        - 4: Very Good
+        - 5: Excellent/Highly Intelligible
+        
+        Note: After each submission, feedback about your guess and current accuracy will appear in the system feedback area.
+        """)
+        
+        # State variables
+        current_participant_id = gr.State(value=generate_participant_id())
+        
+        with gr.Row():
+            with gr.Column():
+                participant_id_display = gr.Textbox(
+                    label="Participant ID",
+                    interactive=False
+                )
+                
+                # Add progress indicator
+                progress_text = gr.Textbox(
+                    label="Progress",
+                    interactive=False,
+                    value="Progress: 0/0 samples"
+                )
+                
+                sample_id = gr.Textbox(label="Sample ID", visible=False)
+                sample_type = gr.Textbox(label="True Sample Type", visible=False)
+                transcription = gr.Textbox(label="Transcription (What should be said)")
+                original_speaker = gr.Textbox(label="Original Speaker", visible=False)
+                synthetic_speaker = gr.Textbox(label="Synthetic Speaker", visible=False)
+                
+                audio_player = gr.Audio(
+                    label="Speech Sample",
+                    type="filepath",
+                    format="wav",
+                    autoplay=False
+                )
+                
+                participant_guess = gr.Radio(
+                    choices=["Yes", "No"],
+                    label="Does this audio sound natural to you?",
+                    value=None
+                )
+                
+                # Declare sliders directly (no gr.Row wrappers)
+                naturalness = gr.Slider(
+                    minimum=1,
+                    maximum=5,
+                    step=1,
+                    value=3,
+                    label="Naturalness Rating",
+                    info="Rate how natural/human-like the speech sounds",
+                    interactive=True
+                )
+                intelligibility = gr.Slider(
+                    minimum=1,
+                    maximum=5,
+                    step=1,
+                    value=3,
+                    label="Intelligibility Rating",
+                    info="Rate how easy it is to understand the speech",
+                    interactive=True
+                )
+                
+                comments = gr.Textbox(
+                    label="Additional Comments",
+                    placeholder="Enter any observations about the speech sample.",
+                    lines=5
+                )
+                
+                # Add a status textbox below comments for feedback
+                status = gr.Textbox(
+                    label="Status / System Feedback",
+                    interactive=False,
+                    lines=2
+                )
+                
+                submit_btn = gr.Button("Submit Rating", variant="primary")
+                next_btn = gr.Button("Next Sample", variant="secondary")
+        
+        def reset_interface():
+            """Reset interface elements to default values"""
+            return {
+                participant_guess: None,
+                naturalness: 3,
+                intelligibility: 3,
+                comments: ""
+            }
+        
+        def load_next_sample(participant_id):
+            """Load the next sample from the playback sequence"""
+            global current_sample_index
+            
+            if current_sample_index >= len(playback_sequence):
+                participant_results = results_df[results_df['participant_id'] == participant_id]
+                final_accuracy = participant_results['guess_correct'].mean() * 100 if not participant_results.empty else 0
+                
+                return [
+                    None,               # audio_player
+                    "Experiment Complete", # sample_id
+                    "Complete",         # sample_type
+                    "",                # transcription
+                    3,                 # naturalness
+                    3,                 # intelligibility
+                    f"Experiment complete! Final accuracy: {final_accuracy:.1f}%", # comments
+                    "Experiment complete!", # status
+                    "",                # original_speaker
+                    "",                # synthetic_speaker
+                    participant_id,    # participant_id_display
+                    None,              # participant_guess
+                    f"Progress: {len(playback_sequence)}/{len(playback_sequence)} samples"  # progress_text
+                ]
+            
+            current_sample = playback_sequence[current_sample_index]
+            
+            # Calculate progress information
+            progress_text = f"Progress: {current_sample_index + 1}/{len(playback_sequence)} samples"
+            
+            # Get sample type and file path
+            sample_type_val = "Original" if current_sample['is_original'] else "Synthetic"
+            audio_file = current_sample['file_path']
+            
+            # Apply noise reduction to the audio file
+            preprocessed_audio = preprocess_audio(audio_file)
+            
+            # Move to next sample
+            current_sample_index += 1
+            
+            return [
+                preprocessed_audio,   # audio_player (now with reduced noise)
+                current_sample['file_name'],  # sample_id
+                sample_type_val,    # sample_type
+                current_sample['transcription'],  # transcription
+                3,                  # naturalness
+                3,                  # intelligibility
+                "",                # comments
+                "",                # status
+                current_sample['original_speaker'],  # original_speaker
+                current_sample['synthetic_speaker'], # synthetic_speaker
+                participant_id,     # participant_id_display
+                None,               # participant_guess
+                progress_text       # progress_text
+            ]
+        
+        def submit_rating(participant_id, sample_id, sample_type, naturalness, intelligibility, comments, 
+                         transcription, original_speaker, synthetic_speaker, participant_guess):
+            """Handle rating submission"""
+            if not participant_guess:
+                return [
+                    gr.skip(),  # audio_player
+                    gr.skip(),  # sample_id
+                    gr.skip(),  # sample_type
+                    gr.skip(),  # transcription
+                    gr.skip(),  # naturalness
+                    gr.skip(),  # intelligibility
+                    gr.skip(),  # comments (do not update)
+                    "Please make a guess before submitting",  # status
+                    gr.skip(),  # original_speaker
+                    gr.skip(),  # synthetic_speaker
+                    gr.skip(),  # participant_id_display
+                    gr.skip(),  # participant_guess
+                    gr.skip()   # progress_text
+                ]
+            
+            # Save and get feedback
+            feedback = save_rating(
+                participant_id, sample_id, sample_type, naturalness, intelligibility, comments,
+                transcription, original_speaker, synthetic_speaker, participant_guess
+            )
+            
+            # Get next sample
+            next_outputs = load_next_sample(participant_id)
+            
+            # The order in load_next_sample is:
+            # [audio_file, file_id, sample_type_val, transcription, naturalness, intelligibility, 
+            #  comments, status, original_speaker, synthetic_speaker, participant_id, participant_guess, progress_text]
+            
+            # Just update the status field (index 7) with the feedback
+            next_outputs[7] = feedback
+            
+            return next_outputs
+        
+        # Event handlers
+        submit_btn.click(
+            submit_rating,
+            inputs=[
+                current_participant_id, sample_id, sample_type, naturalness, intelligibility,
+                comments, transcription, original_speaker, synthetic_speaker, participant_guess
+            ],
+            outputs=[
+                audio_player, sample_id, sample_type, transcription, naturalness,
+                intelligibility, comments, status, original_speaker, synthetic_speaker,
+                participant_id_display, participant_guess, progress_text
+            ]
+        )
+        
+        next_btn.click(
+            load_next_sample,
+            inputs=[current_participant_id],
+            outputs=[
+                audio_player, sample_id, sample_type, transcription, naturalness,
+                intelligibility, comments, status, original_speaker, synthetic_speaker,
+                participant_id_display, participant_guess, progress_text
+            ]
+        )
+        
+        # Load first sample
+        demo.load(
+            load_next_sample,
+            inputs=[current_participant_id],
+            outputs=[
+                audio_player, sample_id, sample_type, transcription, naturalness,
+                intelligibility, comments, status, original_speaker, synthetic_speaker,
+                participant_id_display, participant_guess, progress_text
+            ]
+        )
+    
+    return demo
+
+# Create the interface
+if __name__ == "__main__":
+    demo = create_experiment_interface()
+    demo.launch()

experiment_Results.csv

@@ -0,0 +1 @@
+timestamp,participant_id,sample_id,sample_type,naturalness_rating,intelligibility_rating,comments,transcription,original_speaker,synthetic_speaker,participant_guess,guess_correct

gradio_app.py

@@ -0,0 +1,9 @@
+import gradio as gr
+from experiment1 import create_experiment_interface
+
+# Create the Gradio interface
+demo = create_experiment_interface()
+
+# For Gradio Cloud deployment
+if __name__ == "__main__":
+    demo.launch() 

requirements.txt

@@ -0,0 +1,7 @@
+gradio>=3.50.2
+pandas
+numpy
+soundfile
+librosa
+noisereduce
+uuid 

torgo-synthetic/data/README.md

@@ -0,0 +1,13 @@
+# Synthetic Audio Files
+
+This directory should contain the synthetic audio files for the experiment.
+
+Due to file size constraints, audio files are not included in the git repository but should be uploaded separately.
+
+## Audio File Format
+
+The audio files should be WAV format named according to the pattern in the metadata.csv file.
+
+## Demo Mode
+
+The application will run in demo mode if no audio files are found in this directory. 

torgo-synthetic/metadata.csv

@@ -0,0 +1,99 @@
+file_name,transcription,original_speaker,synthetic_speaker,original_file
+data/F_F03_Session1_0355.wav,air,F03,F03,data/F_F03_Session1_0038.wav
+data/F_F04_Session1_0355.wav,air,F03,F04,data/F_F03_Session1_0038.wav
+data/F_F01_Session1_0355.wav,air,F03,F01,data/F_F03_Session1_0038.wav
+data/F_M01_Session1_0355.wav,air,F03,M01,data/F_F03_Session1_0038.wav
+data/F_M04_Session1_0355.wav,air,F03,M04,data/F_F03_Session1_0038.wav
+data/F_M03_Session1_0355.wav,air,F03,M03,data/F_F03_Session1_0038.wav
+data/F_M02_Session1_0355.wav,air,F03,M02,data/F_F03_Session1_0038.wav
+data/F_F03_Session1_0483.wav,when he speaks his voice is just a bit cracked and quivers a trifle,F03,F03,data/F_F03_Session1_0095.wav
+data/F_F04_Session1_0483.wav,when he speaks his voice is just a bit cracked and quivers a trifle,F03,F04,data/F_F03_Session1_0095.wav
+data/F_F01_Session1_0483.wav,when he speaks his voice is just a bit cracked and quivers a trifle,F03,F01,data/F_F03_Session1_0095.wav
+data/F_M01_Session1_0483.wav,when he speaks his voice is just a bit cracked and quivers a trifle,F03,M01,data/F_F03_Session1_0095.wav
+data/F_M04_Session1_0483.wav,when he speaks his voice is just a bit cracked and quivers a trifle,F03,M04,data/F_F03_Session1_0095.wav
+data/F_M03_Session1_0483.wav,when he speaks his voice is just a bit cracked and quivers a trifle,F03,M03,data/F_F03_Session1_0095.wav
+data/F_M02_Session1_0483.wav,when he speaks his voice is just a bit cracked and quivers a trifle,F03,M02,data/F_F03_Session1_0095.wav
+data/F_F03_Session1_0840.wav,the quick brown fox jumps over the lazy dog,F04,F03,data/F_F04_Session1_0065.wav
+data/F_F04_Session1_0840.wav,the quick brown fox jumps over the lazy dog,F04,F04,data/F_F04_Session1_0065.wav
+data/F_F01_Session1_0840.wav,the quick brown fox jumps over the lazy dog,F04,F01,data/F_F04_Session1_0065.wav
+data/F_M01_Session1_0840.wav,the quick brown fox jumps over the lazy dog,F04,M01,data/F_F04_Session1_0065.wav
+data/F_M04_Session1_0840.wav,the quick brown fox jumps over the lazy dog,F04,M04,data/F_F04_Session1_0065.wav
+data/F_M03_Session1_0840.wav,the quick brown fox jumps over the lazy dog,F04,M03,data/F_F04_Session1_0065.wav
+data/F_M02_Session1_0840.wav,the quick brown fox jumps over the lazy dog,F04,M02,data/F_F04_Session1_0065.wav
+data/F_F03_Session1_0959.wav,knew,F04,F03,data/F_F04_Session1_0008.wav
+data/F_F04_Session1_0959.wav,knew,F04,F04,data/F_F04_Session1_0008.wav
+data/F_F01_Session1_0959.wav,knew,F04,F01,data/F_F04_Session1_0008.wav
+data/F_M01_Session1_0959.wav,knew,F04,M01,data/F_F04_Session1_0008.wav
+data/F_M04_Session1_0959.wav,knew,F04,M04,data/F_F04_Session1_0008.wav
+data/F_M03_Session1_0959.wav,knew,F04,M03,data/F_F04_Session1_0008.wav
+data/F_M02_Session1_0959.wav,knew,F04,M02,data/F_F04_Session1_0008.wav
+data/F_F03_Session1_0973.wav,tear ,F01,F03,data/F_F01_Session1_0007.wav
+data/F_F04_Session1_0973.wav,tear ,F01,F04,data/F_F01_Session1_0007.wav
+data/F_F01_Session1_0973.wav,tear ,F01,F01,data/F_F01_Session1_0007.wav
+data/F_M01_Session1_0973.wav,tear ,F01,M01,data/F_F01_Session1_0007.wav
+data/F_M04_Session1_0973.wav,tear ,F01,M04,data/F_F01_Session1_0007.wav
+data/F_M03_Session1_0973.wav,tear ,F01,M03,data/F_F01_Session1_0007.wav
+data/F_M02_Session1_0973.wav,tear ,F01,M02,data/F_F01_Session1_0007.wav
+data/F_F03_Session1_1071.wav,storm,F01,F03,data/F_F01_Session1_0019.wav
+data/F_F04_Session1_1071.wav,storm,F01,F04,data/F_F01_Session1_0019.wav
+data/F_F01_Session1_1071.wav,storm,F01,F01,data/F_F01_Session1_0019.wav
+data/F_M01_Session1_1071.wav,storm,F01,M01,data/F_F01_Session1_0019.wav
+data/F_M04_Session1_1071.wav,storm,F01,M04,data/F_F01_Session1_0019.wav
+data/F_M03_Session1_1071.wav,storm,F01,M03,data/F_F01_Session1_0019.wav
+data/F_M02_Session1_1071.wav,storm,F01,M02,data/F_F01_Session1_0019.wav
+data/F_F03_Session1_1143.wav,don't ask me to carry an oily rag like that,M01,F03,data/M_M01_Session1_0044.wav
+data/F_F04_Session1_1143.wav,don't ask me to carry an oily rag like that,M01,F04,data/M_M01_Session1_0044.wav
+data/F_F01_Session1_1143.wav,don't ask me to carry an oily rag like that,M01,F01,data/M_M01_Session1_0044.wav
+data/F_M01_Session1_1143.wav,don't ask me to carry an oily rag like that,M01,M01,data/M_M01_Session1_0044.wav
+data/F_M04_Session1_1143.wav,don't ask me to carry an oily rag like that,M01,M04,data/M_M01_Session1_0044.wav
+data/F_M03_Session1_1143.wav,don't ask me to carry an oily rag like that,M01,M03,data/M_M01_Session1_0044.wav
+data/F_M02_Session1_1143.wav,don't ask me to carry an oily rag like that,M01,M02,data/M_M01_Session1_0044.wav
+data/F_F03_Session1_1173.wav,fee,M01,F03,data/M_M01_Session1_0008.wav
+data/F_F04_Session1_1173.wav,fee,M01,F04,data/M_M01_Session1_0008.wav
+data/F_F01_Session1_1173.wav,fee,M01,F01,data/M_M01_Session1_0008.wav
+data/F_M01_Session1_1173.wav,fee,M01,M01,data/M_M01_Session1_0008.wav
+data/F_M04_Session1_1173.wav,fee,M01,M04,data/M_M01_Session1_0008.wav
+data/F_M03_Session1_1173.wav,fee,M01,M03,data/M_M01_Session1_0008.wav
+data/F_M02_Session1_1173.wav,fee,M01,M02,data/M_M01_Session1_0008.wav
+data/F_F03_Session1_1540.wav,both injuries were to the same leg,M04,F03,data/M_M04_Session2_0298.wav
+data/F_F04_Session1_1540.wav,both injuries were to the same leg,M04,F04,data/M_M04_Session2_0298.wav
+data/F_F01_Session1_1540.wav,both injuries were to the same leg,M04,F01,data/M_M04_Session2_0298.wav
+data/F_M01_Session1_1540.wav,both injuries were to the same leg,M04,M01,data/M_M04_Session2_0298.wav
+data/F_M04_Session1_1540.wav,both injuries were to the same leg,M04,M04,data/M_M04_Session2_0298.wav
+data/F_M03_Session1_1540.wav,both injuries were to the same leg,M04,M03,data/M_M04_Session2_0298.wav
+data/F_M02_Session1_1540.wav,both injuries were to the same leg,M04,M02,data/M_M04_Session2_0298.wav
+data/F_F03_Session1_1837.wav,fee,M04,F03,data/M_M04_Session1_0024.wav
+data/F_F04_Session1_1837.wav,fee,M04,F04,data/M_M04_Session1_0024.wav
+data/F_F01_Session1_1837.wav,fee,M04,F01,data/M_M04_Session1_0024.wav
+data/F_M01_Session1_1837.wav,fee,M04,M01,data/M_M04_Session1_0024.wav
+data/F_M04_Session1_1837.wav,fee,M04,M04,data/M_M04_Session1_0024.wav
+data/F_M03_Session1_1837.wav,fee,M04,M03,data/M_M04_Session1_0024.wav
+data/F_M02_Session1_1837.wav,fee,M04,M02,data/M_M04_Session1_0024.wav
+data/F_F03_Session1_1916.wav,know,M03,F03,data/M_M03_Session2_0003.wav
+data/F_F04_Session1_1916.wav,know,M03,F04,data/M_M03_Session2_0003.wav
+data/F_F01_Session1_1916.wav,know,M03,F01,data/M_M03_Session2_0003.wav
+data/F_M01_Session1_1916.wav,know,M03,M01,data/M_M03_Session2_0003.wav
+data/F_M04_Session1_1916.wav,know,M03,M04,data/M_M03_Session2_0003.wav
+data/F_M03_Session1_1916.wav,know,M03,M03,data/M_M03_Session2_0003.wav
+data/F_M02_Session1_1916.wav,know,M03,M02,data/M_M03_Session2_0003.wav
+data/F_F03_Session1_2002.wav,but he always answers banana oil,M03,F03,data/M_M03_Session2_0074.wav
+data/F_F04_Session1_2002.wav,but he always answers banana oil,M03,F04,data/M_M03_Session2_0074.wav
+data/F_F01_Session1_2002.wav,but he always answers banana oil,M03,F01,data/M_M03_Session2_0074.wav
+data/F_M01_Session1_2002.wav,but he always answers banana oil,M03,M01,data/M_M03_Session2_0074.wav
+data/F_M04_Session1_2002.wav,but he always answers banana oil,M03,M04,data/M_M03_Session2_0074.wav
+data/F_M03_Session1_2002.wav,but he always answers banana oil,M03,M03,data/M_M03_Session2_0074.wav
+data/F_M02_Session1_2002.wav,but he always answers banana oil,M03,M02,data/M_M03_Session2_0074.wav
+data/F_F03_Session1_2531.wav,he dresses himself in an ancient black frock coat,M02,F03,data/M_M02_Session1_0044.wav
+data/F_F04_Session1_2531.wav,he dresses himself in an ancient black frock coat,M02,F04,data/M_M02_Session1_0044.wav
+data/F_F01_Session1_2531.wav,he dresses himself in an ancient black frock coat,M02,F01,data/M_M02_Session1_0044.wav
+data/F_M01_Session1_2531.wav,he dresses himself in an ancient black frock coat,M02,M01,data/M_M02_Session1_0044.wav
+data/F_M04_Session1_2531.wav,he dresses himself in an ancient black frock coat,M02,M04,data/M_M02_Session1_0044.wav
+data/F_M03_Session1_2531.wav,he dresses himself in an ancient black frock coat,M02,M03,data/M_M02_Session1_0044.wav
+data/F_M02_Session1_2531.wav,he dresses himself in an ancient black frock coat,M02,M02,data/M_M02_Session1_0044.wav
+data/F_F03_Session1_2631.wav,pat,M02,F03,data/M_M02_Session1_0009.wav
+data/F_F04_Session1_2631.wav,pat,M02,F04,data/M_M02_Session1_0009.wav
+data/F_F01_Session1_2631.wav,pat,M02,F01,data/M_M02_Session1_0009.wav
+data/F_M01_Session1_2631.wav,pat,M02,M01,data/M_M02_Session1_0009.wav
+data/F_M04_Session1_2631.wav,pat,M02,M04,data/M_M02_Session1_0009.wav
+data/F_M03_Session1_2631.wav,pat,M02,M03,data/M_M02_Session1_0009.wav
+data/F_M02_Session1_2631.wav,pat,M02,M02,data/M_M02_Session1_0009.wav

torgo_original/data/README.md

@@ -0,0 +1,13 @@
+# Original Audio Files
+
+This directory should contain the original audio files for the experiment.
+
+Due to file size constraints, audio files are not included in the git repository but should be uploaded separately.
+
+## Audio File Format
+
+The audio files should be WAV format named according to the pattern in the metadata.csv file.
+
+## Demo Mode
+
+The application will run in demo mode if no audio files are found in this directory. 

torgo_original/metadata.csv

@@ -0,0 +1,15 @@
+file_name,transcription
+data/M_M02_Session1_0009.wav,pat
+data/M_M01_Session1_0008.wav,fee
+data/F_F04_Session1_0065.wav,the quick brown fox jumps over the lazy dog
+data/F_F01_Session1_0019.wav,storm
+data/F_F01_Session1_0007.wav,tear 
+data/F_F03_Session1_0095.wav,when he speaks his voice is just a bit cracked and quivers a trifle
+data/M_M03_Session2_0003.wav,know
+data/M_M04_Session1_0024.wav,fee
+data/M_M03_Session2_0074.wav,but he always answers banana oil
+data/M_M02_Session1_0044.wav,he dresses himself in an ancient black frock coat
+data/M_M01_Session1_0044.wav,don't ask me to carry an oily rag like that
+data/F_F03_Session1_0038.wav,air
+data/M_M04_Session2_0298.wav,both injuries were to the same leg
+data/F_F04_Session1_0008.wav,knew