6 accepted papers
Target speaker extraction (TSE) aims to isolate a specific voice from multiple mixed speakers relying on a registerd sample. Since voiceprint features usually vary greatly, current end-to-end neural networks require large model parameters which are computational intensive and impractical for real-ti…
According to the mechanism of speech production, speech can be decomposed into excitation and vocal tract which are sparsely represented in cepstral domain. In this study, we propose a neural network for monaural speech enhancement on time-frequency cepstral space that is implemented by inserting a…
This report summarizes our system submission to the ICASSP 2023 Clarity Challenge. The goal of the challenge is to estimate clean binaural speech signals within a 5 ms system delay. Our submitted Inplace Cepstral Speech Enhancement (ICSE) system features the following aspects: First, we developed a…
Most neural network speech enhancement models ignore speech production mathematical models by directly mapping Fourier transform spectrums or waveforms. In this work, we propose a neural source filter network for speech enhancement. Specifically, we use homomorphic signal processing and cepstral ana…
Recently, deep learning is introduced in acoustic echo cancellation (AEC) and achieves remarkable performance. For deep learning-based AEC, the most important problem is generalization ability in diversity scenarios. Different from most methods which process the entire frequency band, we propose inp…
Under our previous work on frequency bin-wise independent processing, a dramatic reduction of the computational complexity for recurrent neural networks (RNN) is achieved. So that a massive deployment of RNN in time dimension is realized in this paper, by using the channel-wise long short-term memor…