Source Filter

Source-filter models represent signals, particularly audio signals, as the product of an excitation source and a filter shaping its spectral characteristics. Current research focuses on improving these models using neural networks, particularly variational autoencoders and generative adversarial networks (GANs), to achieve better control over signal generation and manipulation, often incorporating techniques like filter pruning for efficiency. This approach has significant implications for applications such as speech enhancement, voice conversion, and sound synthesis, enabling more realistic and controllable audio generation and manipulation.