1 code implementation • 23 Feb 2023 • Shuai Tao, Himavanth Reddy, Jesper Rindom Jensen, Mads Græsbøll Christensen
In addition, the noisy speech and the a posteriori probability SPP representation are used to train our model.
no code implementations • 16 Nov 2022 • Yang Xiang, Jesper Lisby Højvang, Morten Højfeldt Rasmussen, Mads Græsbøll Christensen
To obtain a higher quality enhanced speech, we propose a two-stage DRL-based SE method through adversarial training.
no code implementations • 16 Sep 2022 • Qiongxiu Li, Jaron Skovsted Gundersen, Katrine Tjell, Rafal Wisniewski, Mads Græsbøll Christensen
Privacy has become a major concern in machine learning.
no code implementations • 11 May 2022 • Yang Xiang, Jesper Lisby Højvang, Morten Højfeldt Rasmussen, Mads Græsbøll Christensen
In previous work, we proposed a variational autoencoder-based (VAE) Bayesian permutation training speech enhancement (SE) method (PVAE) which indicated that the SE performance of the traditional deep neural network-based (DNN) method could be improved by deep representation learning (DRL).
no code implementations • 24 Jan 2022 • Yang Xiang, Jesper Lisby Højvang, Morten Højfeldt Rasmussen, Mads Græsbøll Christensen
This means that the proposed method can apply the VAE to model both speech and noise signals, which is totally different from the previous VAE-based SE works.
no code implementations • 4 May 2021 • Alfredo Esquivel Jaramillo, Jesper Kjær Nielsen, Mads Græsbøll Christensen
The segmentation relies on parameter estimates of a hybrid speech model and the maximum a posteriori (MAP) and log-likelihood criteria as rules for model selection among the possible segment lengths, for voiced and unvoiced speech, respectively.
no code implementations • 30 Jun 2020 • Zonglong Bai, Liming Shi, Jinwei Sun, Mads Græsbøll Christensen
In experiments, the proposed algorithm is studied for complex Gaussian random dictionaries and different types of complex signals.
1 code implementation • 29 Apr 2020 • Qiongxiu Li, Richard Heusdens, Mads Græsbøll Christensen
We therefore propose to insert noise in the non-convergent subspace through the dual variable such that the private data are protected, and the accuracy of the desired solution is completely unaffected.
no code implementations • 24 Jun 2017 • Liming Shi, Jesper Kjær Nielsen, Jesper Rindom Jensen, Mads Græsbøll Christensen
The modeling of speech can be used for speech synthesis and speech recognition.