A new speech coding paper, “Cascaded Cross-Module Residual Learning towards Lightweight End-to-End Speech Coding,” was accepted for publication in Interspeech 2019.

The Ph.D. students in SAIGE work as an intern in various research labs in the industry this summer: Amazon (Sanna), LinkedIn (Kai), Qualcomm (Sunwoo), and Spotify (Aswin).

We are attending ICASSP 2019 in Brighten, UK. Minje will chair a session on source separation and speech enhancement; Sunwoo and Sanna will present their papers about bitwise recurrent neural networks and a database of quality Karaoke singing, respectively.

The Deep Autotuner project got featured on BBC Radio 5 Live “Drive (from 1:27:09),”The Times, Daily Mail, and the New Scientist magazine. It also got featured on School’s news page.

Our papers are accepted for publication in ICASSP 2019: “Incremental Binarization On Recurrent Neural Networks for Single-Channel Source Separation” and “Intonation: a Dataset of Quality Vocal Performances Refined by Spectral Clustering on Pitch Congruence.”

A fun collaboration with folks in the Sunflower state resulted in a paper in RTCSA. It’s about DeepPicar, a deep learning based low powered autonomous driving system: “DeepPicar: A Low-cost Deep Neural Network-based Autonomous Car”

Our paper is accepted for publication in EUSIPCO 2018: Sanna Wager and Minje Kim, “Collaborative speech dereverberation: regularized tensor factorization for crowdsourced multi-channel recordings.”

Kai passed his qualifying exam. Congrats!

We’ve got two ICASSP 2018 papers accepted:“Bitwise Source Separation on Hashed Spectra: An Efficient Posterior Estimation Scheme Using Partial Rank Order Metrics,” and“Bitwise Neural Networks for Efficient Single-Channel Source Separation.”

At NIPS 2017 Workshop on Machine Learning for Audio, SAIGE presented two posters about efficient speech enhancement algorithms using bitwise machine learning models.