Spaces:
Running
Running
| import streamlit as st | |
| import tensorflow as tf | |
| import numpy as np | |
| import librosa | |
| import matplotlib.pyplot as plt | |
| import librosa.display | |
| import tempfile | |
| import os | |
| # Load the trained model | |
| def load_model(): | |
| model_path = "sound_classification_model.h5" # Replace with the path to your .h5 file | |
| model = tf.keras.models.load_model(model_path) | |
| return model | |
| model = load_model() | |
| # Preprocess audio into a spectrogram | |
| def preprocess_audio(file_path, n_mels=128, fixed_time_steps=128): | |
| try: | |
| y, sr = librosa.load(file_path, sr=None) | |
| mel_spectrogram = librosa.feature.melspectrogram(y=y, sr=sr, n_mels=n_mels, fmax=sr/2) | |
| log_spectrogram = librosa.power_to_db(mel_spectrogram, ref=np.max) | |
| log_spectrogram = log_spectrogram / np.max(np.abs(log_spectrogram)) | |
| if log_spectrogram.shape[1] < fixed_time_steps: | |
| padding = fixed_time_steps - log_spectrogram.shape[1] | |
| log_spectrogram = np.pad(log_spectrogram, ((0, 0), (0, padding)), mode='constant') | |
| else: | |
| log_spectrogram = log_spectrogram[:, :fixed_time_steps] | |
| return np.expand_dims(log_spectrogram, axis=-1) # Add channel dimension for CNNs | |
| except Exception as e: | |
| print(f"Error processing {file_path}: {e}") | |
| return None | |
| # Streamlit app UI | |
| st.title("Audio Spectrogram Prediction") | |
| st.write("Upload an audio file to generate a spectrogram and predict its class using your trained model.") | |
| # File upload widget | |
| uploaded_file = st.file_uploader("Choose an audio file", type=["wav", "mp3"]) | |
| if uploaded_file is not None: | |
| # Save the uploaded audio file to a temporary location | |
| with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as temp_audio_file: | |
| temp_audio_file.write(uploaded_file.read()) | |
| temp_audio_path = temp_audio_file.name | |
| # Preprocess the audio into a spectrogram | |
| st.write("Processing audio into a spectrogram...") | |
| spectrogram = preprocess_audio(temp_audio_path) | |
| if spectrogram is not None: | |
| # Display the spectrogram | |
| st.write("Generated Spectrogram:") | |
| plt.figure(figsize=(10, 4)) | |
| librosa.display.specshow(spectrogram[:, :, 0], sr=22050, x_axis='time', y_axis='mel', fmax=8000, cmap='plasma') | |
| plt.colorbar(format='%+2.0f dB') | |
| plt.title('Mel-Spectrogram') | |
| plt.tight_layout() | |
| st.pyplot(plt) | |
| # Predict using the model | |
| st.write("Predicting...") | |
| spectrogram = np.expand_dims(spectrogram, axis=0) # Add batch dimension | |
| predictions = model.predict(spectrogram) | |
| predicted_class = np.argmax(predictions, axis=-1) # Assuming classification | |
| # Display the results | |
| st.write("Prediction Results:") | |
| st.write(f"Predicted Class: {int(predicted_class[0])}") | |
| st.write(f"Raw Model Output: {predictions}") | |
| else: | |
| st.write("Failed to process the audio file. Please try again with a different file.") | |
| # Optional: Clean up temporary file | |
| os.remove(temp_audio_path) | |