Yun-Ning Hung

Title: Yun-Ning Hung: Advancing Automatic Music Transcription

Introduction

Yun-Ning Hung is an inventor based in Culver City, California, and is associated with Lemon Inc. Her patent work focuses on applying deep neural networks to automatic music audio transcription.

Latest Patents

Hung is named on 1 patent, titled “Implementing Automatic Music Audio Transcription.” The disclosure describes a deep neural network that processes spectral representations of audio over time. Its components include a spectral cross-attention sub-model that projects each time step’s spectral representation into latent arrays, latent transformers that perform self-attention on those arrays, and temporal transformers that enable communication between arrays at different time steps.

The patent also describes augmenting training data by randomly mixing datasets, including vocal and instrumental datasets, and training the model on the resulting data. These techniques are intended to support automatic transcription of music audio.

Career Highlights

Hung’s patent reflects work at the intersection of machine learning and music technology. It outlines a model architecture designed to capture both spectral features within individual time steps and relationships across time.

Collaborations

The patent lists Ju-Chiang Wang and Wei Tsung Lu as coworkers, indicating collaboration on the disclosed music-transcription technology.

Conclusion

Yun-Ning Hung’s patent contribution centers on a neural-network approach to automatic music transcription, combining spectral and temporal processing with mixed vocal and instrumental training data.