Dengfeng Ke

About

Dengfeng Ke is a researcher whose work lies at the intersection of deep learning and automatic speech recognition (ASR). His key research areas include recurrent neural network architectures, particularly Long Short-Term Memory (LSTM) models, and efficient training methods for multilayer perceptrons (MLPs) in large-scale speech systems. Ke’s most cited paper, "Long Short-Term Memory Projection Recurrent Neural Network Architectures for Piano’s Continuous Note Recognition" (2017, 31 citations), introduces an LSTMP variant that optimizes both speed and performance for time-series tasks, with notable application to music note recognition. His earlier work, "A Fast Learning Method for Multilayer Perceptrons in Automatic Speech Recognition Systems" (2015, 14 citations), proposes a preadjusting strategy using data separation and a cosine-based dynamic learning rate to accelerate MLP training for large vocabulary continuous speech recognition (LVCSR). These contributions demonstrate Ke’s focus on improving neural network efficiency and accuracy in real-world audio processing, making his research valuable for students and engineers working on speech and music recognition systems.

Research Focus

Key Achievements

Top Papers

1. 1
   Long Short-Term Memory Projection Recurrent Neural Network Architectures for Piano’s Continuous Note Recognition

   31 citations · 2017
2. 2
   A Fast Learning Method for Multilayer Perceptrons in Automatic Speech Recognition Systems

   14 citations · 2015

Key Collaborators

Contact & Links