The Vocal Denoiser Plugin is an advanced audio processing tool designed to enhance speech clarity by effectively reducing background noise in real time. This plugin utilizes a convolutional neural network (CNN) to suppress unwanted noise while preserving speech integrity.
- Real-Time Noise Suppression: Uses a pre-trained deep learning model to analyze and remove noise from speech.
- Short-Time Fourier Transform (STFT): Converts the time-domain signal into a frequency-domain representation for precise processing.
- Voice Activity Detection (VAD): Identifies and isolates speech from non-speech regions, ensuring that the noise suppression is applied only where necessary.
- Adaptive Noise Gate: Dynamically adjusts threshold levels based on detected speech probability to further refine noise reduction.
- Multiple Noise Profiles: Users can select from predefined noise profiles:
- Washing Machine
- White Noise
- Crowd Noise
- Sample Rate Conversion: Supports multiple sample rates and automatically converts signals to ensure compatibility with various audio workflows.
- User Customization: Adjustable Strength parameter allows fine-tuning of the denoising effect.
The plugin processes incoming audio through the following stages:
- Noise Profile Selection: Users select a noise type that best matches their environment.
- Preprocessing with STFT: The signal is transformed into its frequency representation using Short-Time Fourier Transform (STFT).
- Neural Network Denoising: A CNN trained on noisy and clean speech pairs processes the STFT data, estimating a clean speech output.
- Blending Mechanism: The denoised output is blended with the original signal based on user-defined strength levels to maintain natural speech quality.
- VAD Integration: The VAD model detects speech segments, ensuring that the noise suppression is applied only when necessary.
- Final Reconstruction: The processed signal is converted back to the time domain using the Inverse Short-Time Fourier Transform (ISTFT) and outputted at the original sample rate.