Company Filing History:
Years Active: 2021-2024
Title: Innovations of Inventor Zhao Song
Introduction
Zhao Song is a prominent inventor based in Sunnyvale, California. He has made significant contributions to the field of audio processing, holding a total of three patents. His work focuses on advanced techniques for audio denoising and generative models.
Latest Patents
One of Zhao Song's latest patents is titled "Speech denoising via discrete representation learning." This invention presents a new end-to-end approach for audio denoising, synthesizing denoised audio directly from a generative model. Instead of modeling the noise component explicitly, this method generates phonetic content using a variational autoencoder with discrete latent representations. The invention also introduces a new matching loss for denoising, achieving competitive performance compared to other methods.
Another notable patent is "Small-footprint flow-based models for raw audio," which introduces WaveFlow, a generative flow model for raw audio. WaveFlow utilizes a dilated two-dimensional convolutional architecture to handle long-range structures while modeling local variations with autoregressive functions. This model generates high-fidelity speech significantly faster than existing systems, with a small footprint of only 5.91 million parameters.
Career Highlights
Zhao Song is currently employed at Baidu USA LLC, where he continues to innovate in the field of audio technology. His work has garnered attention for its efficiency and effectiveness in audio synthesis.
Collaborations
Zhao has collaborated with notable colleagues, including Wei Ping and Kainan Peng, contributing to advancements in audio processing technologies.
Conclusion
Zhao Song's innovative patents and contributions to audio technology highlight his expertise and commitment to advancing the field. His work continues to influence the development of efficient audio processing methods.