An Alternative Noise Cancelling Approach by Anti-Noise Generation Model Based on Diffusion^*

These days to have audio equipment without a noise-canceling function is unimaginable. This feature has become an integral part of recent audio devices. Consequently, noise-canceling technologies have made significant advancements. However, many of these technologies can cause auditory discomfort due to an incomplete match between the detected noise and the generated anti-noise. In response to these challenges, this paper introduces an alternative noise-canceling approach using an anti-noise generation model. Unlike previous algorithm-based noise-canceling methods, we constructed a Transformer-based diffusion model that directly generates a combination of source music and anti-noise. Traditional methods are more complex and have certain limitations. However, our model detects noise in latent space and uses the original source music as conditional information. Additionally, we have secured datasets that allow the model to learn and adapt to various situations. These elements contribute to more sophisticated and precise noise-canceling performance.