Spaces:
Running
Running
| """Set up the Gradio interface""" | |
| import gradio as gr | |
| from transformers import pipeline | |
| from TTS.api import TTS | |
| # Load pre-trained emotion detection model | |
| emotion_classifier = pipeline("text-classification", model="bhadresh-savani/distilbert-base-uncased-emotion") | |
| # Load TTS model | |
| tts_model = TTS(model_name="tts_models/en/ljspeech/tacotron2-DDC") | |
| # Emotion-specific settings for pitch and speed | |
| emotion_settings = { | |
| "neutral": {"pitch": 1.0, "speed": 1.0}, | |
| "joy": {"pitch": 1.3, "speed": 1.2}, | |
| "sadness": {"pitch": 0.8, "speed": 0.9}, | |
| "anger": {"pitch": 1.6, "speed": 1.4}, | |
| "fear": {"pitch": 1.2, "speed": 0.95}, | |
| "surprise": {"pitch": 1.5, "speed": 1.3}, | |
| "disgust": {"pitch": 0.9, "speed": 0.95}, | |
| "shame": {"pitch": 0.8, "speed": 0.85}, | |
| } | |
| import librosa | |
| import soundfile as sf | |
| def adjust_audio_speed(audio_path, speed_factor): | |
| y, sr = librosa.load(audio_path) | |
| y_speeded = librosa.effects.time_stretch(y, speed_factor) | |
| sf.write(audio_path, y_speeded, sr) | |
| def adjust_audio_pitch(audio_path, pitch_factor): | |
| y, sr = librosa.load(audio_path) | |
| y_shifted = librosa.effects.pitch_shift(y, sr, n_steps=pitch_factor) | |
| sf.write(audio_path, y_shifted, sr) | |
| def emotion_aware_tts_pipeline(input_text=None, file_input=None): | |
| try: | |
| # Get text from input or file | |
| if file_input: | |
| with open(file_input.name, 'r') as file: | |
| input_text = file.read() | |
| if input_text: | |
| # Detect emotion | |
| emotion_data = emotion_classifier(input_text)[0] | |
| emotion = emotion_data['label'] | |
| confidence = emotion_data['score'] | |
| # Adjust pitch and speed | |
| settings = emotion_settings.get(emotion.lower(), {"pitch": 1.0, "speed": 1.0}) | |
| pitch = settings["pitch"] | |
| speed = settings["speed"] | |
| # Generate audio | |
| audio_path = "output.wav" | |
| tts_model.tts_to_file(text=input_text, file_path=audio_path) | |
| # Adjust pitch and speed using librosa | |
| if pitch != 1.0: | |
| adjust_audio_pitch(audio_path, pitch) | |
| if speed != 1.0: | |
| adjust_audio_speed(audio_path, speed) | |
| return f"Detected Emotion: {emotion} (Confidence: {confidence:.2f})", audio_path | |
| else: | |
| return "Please provide input text or file", None | |
| except Exception as e: | |
| return f"Error: {str(e)}", None | |
| # Define Gradio interface | |
| iface = gr.Interface( | |
| fn=emotion_aware_tts_pipeline, | |
| inputs=[ | |
| gr.Textbox(label="Input Text", placeholder="Enter text here"), | |
| gr.File(label="Upload a Text File") | |
| ], | |
| outputs=[ | |
| gr.Textbox(label="Detected Emotion"), | |
| gr.Audio(label="Generated Audio") | |
| ], | |
| title="Emotion-Aware Text-to-Speech", | |
| description="Input text or upload a text file to detect the emotion and generate audio with emotion-aware modulation." | |
| ) | |
| # Launch Gradio interface | |
| iface.launch() |