Multi-distribution deep belief network for speech synthesis

By Shiyin Kang, Xiaojun Qian, Helen Mei-Ling Meng. 2013

110 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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

1. Junichi Yamagishi
2. Frank K. Soong
3. Dong Yu
4. Zhizheng Wu
5. Li Deng
6. Zhenhua Ling
7. Simon King
8. Xu Tan

In Signal Processing

1. Junichi Yamagishi
2. Dong Yu
3. Li Deng
4. Zhizheng Wu
5. Frank K. Soong
6. Simon King
7. Jinyu Li
8. Zhenhua Ling

Citations in the corpus

• Deep Learning: Methods and Applications (2014, methodology): “In the approach developed in [190], the DBN as a generative model is used to represent joint distribution of linguistic and acoustic features.”
• A tutorial survey of architectures, algorithms, and applications for deep learning (2014, methodology): “In the approach developed in [182], the DBN as a generative model is used to represent joint distribution of linguistic and acoustic features.”
• A Deep Generative Architecture for Postfiltering in Statistical Parametric Speech Synthesis (2015, methodology): “DBNs have also been used to model the joint distribution of linguistic and acoustic features [7].”
• TTS synthesis with bidirectional LSTM based recurrent neural networks (2014, methodology): “…in the input data as generative “pre - training” and find a region of the weight-space that can reduce over-fitting for the discriminative “fine - tuning” phase in speech recognition [ 6], is also employed to model joint distribution of linguistic and acoustic features for speech synthesis [7].”
• Deep neural networks employing Multi-Task Learning and stacked bottleneck features for speech synthesis (2015, methodology): “The second problem – frame-by-frame independence – arises in many SPSS systems where predicted values for consecutive acoustic states or frames are conditionally independent of one another given their linguistic contexts.”
• Statistical parametric speech synthesis using generative adversarial networks under a multi-task learning framework (2017, unknown): “More recently, advanced estimation criteria and novel network architectures have been introduced to further improve the performance of SPSS [7, 8, 9, 10, 11].”
• Merlin: An Open Source Neural Network Speech Synthesis System (2016, methodology): “In [13, 14], deep belief networks (DBNs) as deep generative model were employed to model the relationship between linguistic and acoustic features jointly.”
• 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].”
• Voice Conversion Using Deep Neural Networks With Layer-Wise Generative Training (2014, methodology): “The proposed four-layer DNN is trained layer-by-layer from a cas­cade of a Bernoulli BAM (BBAM) and two RBMs.”
• 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”
• Modeling Spectral Envelopes Using Restricted Boltzmann Machines and Deep Belief Networks for Statistical Parametric Speech Synthesis (2013, methodology): “Similar approach for the statistical parametric speech synthesis has been studied in [30], where the joint distribution between the tonal syllable ID and the spectral and excitation features is modeled using a multi-distribution DBN.”
• 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 Learning in Spiking Neural Networks (2018, methodology): “DBNs and convolutional DBNs have successfully been employed in many areas such as visual processing [192], [193], [194], [195], audio processing [196], [197], [198], [199], [200], time series forecasting [201], and protein folding [202].”
• A survey on speech synthesis techniques in Indian languages (2020, methodology): “The char2wave (https ://mila.quebe c/en/publi catio n/char2 wav-end-to-end-speec h-synth esis/) is an end-to-end model for speech synthesis that uses the Recurrent Neural Network (RNN) concept to produce acoustic features for speech production.”
• A Review of Deep Learning Based Speech Synthesis (2019, methodology): “The multi-distribution deep belief network (DBN) [43] is a method of modeling the joint distribution of context information and acoustic features.”