D
Dong Yu
Researcher at Tencent
Publications - 389
Citations - 45733
Dong Yu is an academic researcher from Tencent. The author has contributed to research in topics: Artificial neural network & Word error rate. The author has an hindex of 72, co-authored 339 publications receiving 39098 citations. Previous affiliations of Dong Yu include Peking University & Microsoft.
Papers
More filters
Proceedings ArticleDOI
FastDiff: A Fast Conditional Diffusion Model for High-Quality Speech Synthesis
TL;DR: FastDiff is proposed, a fast conditional diffusion model for high-quality speech synthesis that employs a stack of time-aware location-variable convolutions of diverse receptive field patterns to efficiently model long-term time dependencies with adaptive conditions and generalized well to the mel-spectrogram inversion of unseen speakers.
Journal ArticleDOI
A Novel Framework and Training Algorithm for Variable-Parameter Hidden Markov Models
TL;DR: It is shown that under the well-matched condition the proposed discriminatively trained VPHMM outperforms the conventional HMM trained in the same way with relative word error rate (WER) reduction of 19% and 15%, respectively, when only mean is updated and when both mean and variances are updated.
Proceedings ArticleDOI
Deep-structured hidden conditional random fields for phonetic recognition.
TL;DR: Both the DHCRF and the HMM are superior to the discriminatively trained tri-phone hidden Markov model using identical input features and the use of this new sequential deep-learning model for phonetic recognition is investigated.
Proceedings ArticleDOI
Speaker-aware training of LSTM-RNNS for acoustic modelling
TL;DR: This paper studies the LSTM-RNN speaker-aware training that incorporates the speaker information during model training to normalise the speaker variability, and empirically evaluates three types of speaker representation: I-vectors, bottleneck speaker vectors and speaking rate.
Patent
Exploiting sparseness in training deep neural networks
TL;DR: In this paper, the sparseness of non-zero hidden layer interconnection weight values is exploited to train a fully connected DNN by sweeping through a full training set a number of times and only the interconnections whose weight magnitudes exceed a minimum weight threshold are considered in further training.