Upload 5 files

README.md

@@ -1,14 +1 @@
----
-title: VocalBiomarkersForMentalHealth
-emoji: 🌍
-colorFrom: pink
-colorTo: pink
-sdk: gradio
-sdk_version: 5.6.0
-app_file: app.py
-pinned: false
-license: mit
-short_description: AI powered Vocal Biomarkers for Mental Health
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+This repository is empty. Any task executed on it should be considered a new capability and needs to be created.

app.py

@@ -0,0 +1,42 @@
+import gradio as gr
+from transformers import pipeline
+import librosa
+import numpy as np
+import torch
+
+# Load pre-trained models from Hugging Face Hub
+emotion_model = pipeline("sentiment-analysis", model="bhadresh-savani/distilbert-base-uncased-emotion")
+
+# Define preprocessing functions for vocal input data
+def preprocess_audio(audio):
+    y, sr = librosa.load(audio, sr=16000)
+    return y, sr
+
+def extract_features(y, sr):
+    features = {
+        "mfcc": librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13).mean(axis=1),
+        "chroma": librosa.feature.chroma_stft(y=y, sr=sr).mean(axis=1),
+        "mel": librosa.feature.melspectrogram(y=y, sr=sr).mean(axis=1),
+        "contrast": librosa.feature.spectral_contrast(y=y, sr=sr).mean(axis=1),
+        "tonnetz": librosa.feature.tonnetz(y=y, sr=sr).mean(axis=1)
+    }
+    return features
+
+# Define prediction functions to analyze vocal biomarkers
+def analyze_emotion(audio):
+    y, sr = preprocess_audio(audio)
+    features = extract_features(y, sr)
+    input_data = torch.tensor([features["mfcc"], features["chroma"], features["mel"], features["contrast"], features["tonnetz"]])
+    result = emotion_model(input_data)
+    return result
+
+# Create Gradio interface for user input and analysis results
+def analyze(audio):
+    result = analyze_emotion(audio)
+    return result
+
+iface = gr.Interface(fn=analyze, inputs=gr.inputs.Audio(source="microphone", type="filepath"), outputs="text")
+
+# Launch the Gradio app
+if __name__ == "__main__":
+    iface.launch()

data_preprocessing.py

@@ -0,0 +1,25 @@
+import librosa
+import numpy as np
+import scipy
+
+def preprocess_audio(audio):
+    y, sr = librosa.load(audio, sr=16000)
+    return y, sr
+
+def clean_audio(y, sr):
+    y = librosa.effects.trim(y)[0]
+    y = librosa.util.normalize(y)
+    return y
+
+def extract_features(y, sr):
+    features = {
+        "mfcc": librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13).mean(axis=1),
+        "chroma": librosa.feature.chroma_stft(y=y, sr=sr).mean(axis=1),
+        "mel": librosa.feature.melspectrogram(y=y, sr=sr).mean(axis=1),
+        "contrast": librosa.feature.spectral_contrast(y=y, sr=sr).mean(axis=1),
+        "tonnetz": librosa.feature.tonnetz(y=y, sr=sr).mean(axis=1)
+    }
+    return features
+
+def store_preprocessed_data(features, filename):
+    np.savez(filename, **features)

model_training.py

@@ -0,0 +1,70 @@
+import torch
+from transformers import AutoModelForSequenceClassification, AutoTokenizer
+from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, roc_auc_score, confusion_matrix, mean_absolute_error, mean_squared_error
+from data_preprocessing import preprocess_audio, extract_features
+
+# Load pre-trained models from Hugging Face Hub
+model_name = "bhadresh-savani/distilbert-base-uncased-emotion"
+model = AutoModelForSequenceClassification.from_pretrained(model_name)
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+# Prepare the dataset using functions from data_preprocessing.py
+def prepare_dataset(audio_files):
+    features = []
+    labels = []
+    for audio in audio_files:
+        y, sr = preprocess_audio(audio)
+        feature = extract_features(y, sr)
+        features.append(feature)
+        # Assuming labels are provided in the filename as the last character before the extension
+        label = int(audio.split('_')[-1].split('.')[0])
+        labels.append(label)
+    return features, labels
+
+# Define the training loop for fine-tuning models
+def train_model(model, tokenizer, train_features, train_labels, epochs=3, batch_size=8):
+    optimizer = torch.optim.AdamW(model.parameters(), lr=5e-5)
+    model.train()
+    for epoch in range(epochs):
+        for i in range(0, len(train_features), batch_size):
+            batch_features = train_features[i:i+batch_size]
+            batch_labels = train_labels[i:i+batch_size]
+            inputs = tokenizer(batch_features, return_tensors="pt", padding=True, truncation=True)
+            labels = torch.tensor(batch_labels)
+            outputs = model(**inputs, labels=labels)
+            loss = outputs.loss
+            loss.backward()
+            optimizer.step()
+            optimizer.zero_grad()
+            print(f"Epoch {epoch+1}, Batch {i//batch_size+1}, Loss: {loss.item()}")
+
+# Implement functions to evaluate model performance
+def evaluate_model(model, tokenizer, test_features, test_labels):
+    model.eval()
+    predictions = []
+    with torch.no_grad():
+        for feature in test_features:
+            inputs = tokenizer(feature, return_tensors="pt", padding=True, truncation=True)
+            outputs = model(**inputs)
+            prediction = torch.argmax(outputs.logits, dim=1).item()
+            predictions.append(prediction)
+    accuracy = accuracy_score(test_labels, predictions)
+    precision = precision_score(test_labels, predictions, average='weighted')
+    recall = recall_score(test_labels, predictions, average='weighted')
+    f1 = f1_score(test_labels, predictions, average='weighted')
+    auc_roc = roc_auc_score(test_labels, predictions, average='weighted', multi_class='ovr')
+    conf_matrix = confusion_matrix(test_labels, predictions)
+    mae = mean_absolute_error(test_labels, predictions)
+    rmse = mean_squared_error(test_labels, predictions, squared=False)
+    return accuracy, precision, recall, f1, auc_roc, conf_matrix, mae, rmse
+
+# Save the trained models in a usable format
+def save_model(model, path):
+    model.save_pretrained(path)
+    tokenizer.save_pretrained(path)
+
+# Log training progress and performance metrics
+def log_metrics(metrics, filename):
+    with open(filename, 'w') as f:
+        for key, value in metrics.items():
+            f.write(f"{key}: {value}\n")

requirements.txt

@@ -0,0 +1,8 @@
+gradio
+transformers
+torch
+librosa
+numpy
+scipy
+pandas
+scikit-learn