Modeling Spectral Envelopes Using Restricted Boltzmann Machines and Deep Belief Networks for Statistical Parametric Speech Synthesis

By Zhenhua Ling, Li Deng, Dong Yu. 2013

164 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Generative Models, Speech & Audio, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Generative Models

In Speech & Audio

In Signal Processing

Citations in the corpus

Deep Learning: Methods and Applications (2014, methodology): “[227, 229], the RBM and DBN as generative models are used to replace the traditional Gaussian models, achieving significant quality improvement, in both subjective and objective measures, of the synthesized voice.”
Forecasting Crude Oil Prices: a Deep Learning based Model (2017, methodology): “For example, [7] proposed HMM-DNN model for speech synthesis method.”
Fine-Grained Emotion Strength Transfer, Control and Prediction for Emotional Speech Synthesis (2020, background): “Thanks to the rapid development of deep learning, speech synthesis has been significantly advanced [1, 2, 3].”
A tutorial survey of architectures, algorithms, and applications for deep learning (2014, methodology): “[64, 181], the RBM and DBN as generative models are used to replace the traditional Gaussian models, achieving significant quality improvement, in both subjective and objective measures, of the synthesized voice.”
DNN-based stochastic postfilter for HMM-based speech synthesis (2014, methodology): “, deep belief networks (DBNs) [11, 12] or deep Boltzmann machines (DBMs) [13], to construct deeper mapping relationships.”
A Deep Generative Architecture for Postfiltering in Statistical Parametric Speech Synthesis (2015, methodology): “Therefore, the DBN has a better ability to describe the probabilistic distribution of visible variables than the RBM [6], [28].”
Spoofing and countermeasures for speaker verification: A survey (2015, background): “In addition to the four major approaches, inspired by advances in deep neural network (DNN)-based speech recognition (Hinton et al., 2012), new data-driven, DNN-based approaches have also been actively investigated (Zen et al., 2013; Ling et al., 2013; Lu et al., 2013; Qian et al., 2014).”
Modeling and Transforming Speech Using Variational Autoencoders (2016, methodology): “These models have been used for tasks such as speech recognition [7], speech synthesis [8], and other tasks like voice conversion [9].”
Implementing Prosodic Phrasing in Chinese End-to-end Speech Synthesis (2019, methodology): “The statistical parametric methods, either HMM-based [3][4] or DNN-based [5][6][7][8], build acoustic models for the language and generate a sequence of acoustic parameters, which are then converted into speech signals by a vocoder.”
Deep neural networks employing Multi-Task Learning and stacked bottleneck features for speech synthesis (2015, methodology): “In [9], restricted Boltzmann machines (RBMs) were used to replace Gaussian mixture models over the acoustic features, allowing more spectral detail to be learned, which resulted in better speech quality.”
A deep learning approach for credit scoring using credit default swaps (2017, background): “…with success in classification task, e.g., computer vision (Krizhevsky et al., 2012), health state classification (Tamilselvan and Wang, 2013; Abdel-Zaher and Eldeib, 2016), speech and language processing (Mohamed et al., 2012; Ling et al., 2013) and emotion recognition (Le and Provost, 2013).”
A deep auto-encoder based low-dimensional feature extraction from FFT spectral envelopes for statistical parametric speech synthesis (2016, methodology): “Restricted Boltzmann machines or deep belief networks have been used to model the output probabilities of hidden Markov model (HMM) states instead of Gaussian mixture models (GMMs) [2].”
Machine Learning for Synthetic Data Generation: a Review (2023, background): “The proposed conventional hidden Markov model (HMM)-based speech synthesis system can be significantly improved in terms of naturalness and over-smoothing [61].”
Deep mixture density networks for acoustic modeling in statistical parametric speech synthesis (2014, background): “There have been many attempts to improve the accuracy, such as trajectory HMMs [4], autoregressive HMMs [5], minimum generation error (MGE) training [6], product of experts (PoEs) [7,8], Gaussian process regression (GPR) [9], and restricted Boltzmann machines (RBMs) [10, 11].”
A study of speaker adaptation for DNN-based speech synthesis (2015, background): “Several independent studies have shown that DNNs can produce more natural synthesised speech than the conventional HMM-based speech synthesis for a single speaker in various training conditions [7, 8, 9, 10, 11, 12, 13, 14], but only few studies have addressed the question of whether DNN-based speec”