Spaces:
Running
Running
| import gradio as gr | |
| from transformers import pipeline | |
| import librosa | |
| import numpy as np | |
| import soundfile as sf | |
| import os | |
| # --- Model Loading --- | |
| # We'll use the pipeline abstraction from transformers for simplicity. | |
| # This model is specifically designed for audio classification (emotion detection). | |
| # It will automatically handle the loading of the model and its preprocessor. | |
| classifier = pipeline("audio-classification", model="mrm8488/Emotion-detection-from-audio-files") | |
| # --- Emotion Labels Mapping (Optional, for clearer output) --- | |
| # The model outputs raw labels, we can define a more readable mapping if needed | |
| # For this specific model, the labels are already pretty clear. | |
| # Example labels from the model's page: 'anger', 'disgust', 'fear', 'happiness', 'neutral', 'sadness', 'surprise' | |
| # --- Prediction Function --- | |
| def predict_emotion(audio_file): | |
| """ | |
| Predicts emotions from an audio file. | |
| Args: | |
| audio_file (str or np.ndarray): Path to the audio file or a numpy array | |
| (if using microphone input directly). | |
| Gradio's Audio component usually provides | |
| a file path for file uploads or a tuple | |
| (samplerate, audio_array) for microphone. | |
| Returns: | |
| dict: A dictionary of emotion labels and their probabilities. | |
| """ | |
| if audio_file is None: | |
| return {"error": "No audio input provided."} | |
| # Gradio's Audio component can return a path to a temp file for file uploads, | |
| # or a tuple (samplerate, numpy_array) for microphone input. | |
| if isinstance(audio_file, str): | |
| # Handle file path (e.g., from file upload) | |
| audio_path = audio_file | |
| elif isinstance(audio_file, tuple): | |
| # Handle microphone input (samplerate, numpy_array) | |
| sample_rate, audio_array = audio_file | |
| # Save the numpy array to a temporary WAV file as the pipeline expects a file path or direct bytes | |
| temp_audio_path = "temp_audio_from_mic.wav" | |
| sf.write(temp_audio_path, audio_array, sample_rate) | |
| audio_path = temp_audio_path | |
| else: | |
| return {"error": "Invalid audio input format."} | |
| try: | |
| # Perform inference | |
| results = classifier(audio_path) | |
| # Process results into a dictionary for better display | |
| emotion_scores = {item['label']: item['score'] for item in results} | |
| return emotion_scores | |
| except Exception as e: | |
| return {"error": f"An error occurred during prediction: {str(e)}"} | |
| finally: | |
| # Clean up temporary file if created | |
| if 'temp_audio_path' in locals() and os.path.exists(temp_audio_path): | |
| os.remove(temp_audio_path) | |
| # --- Gradio Interface --- | |
| # Define the Gradio interface | |
| iface = gr.Interface( | |
| fn=predict_emotion, | |
| inputs=gr.Audio(type="filepath", label="Upload Audio or Record with Microphone", sources=["microphone", "file"]), | |
| outputs=gr.Label(num_top_classes=7, label="Emotion Probabilities"), # Adjust num_top_classes based on model's output labels | |
| title="AI Audio Emotion Detector", | |
| description="Upload an audio file or record your voice to detect emotions like anger, disgust, fear, happiness, neutral, sadness, and surprise." | |
| ) | |
| # Launch the Gradio app | |
| if __name__ == "__main__": | |
| iface.launch() |