Deep mixture density networks for acoustic modeling in statistical parametric speech synthesis

By Heiga Zen, Andrew W. Senior. 2014

235 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

In Signal Processing

Citations in the corpus

Fast, Compact, and High Quality LSTM-RNN Based Statistical Parametric Speech Synthesizers for Mobile Devices (2016, background): “ANN-based acoustic models offer an efﬁcient and distributed representation of complex dependencies between linguistic and acoustic features [21, 22] and have shown the potential to produce natural sounding synthesized speech [2, 4, 7–9].”
Revisiting Over-Smoothness in Text to Speech (2022, background): “Therefore, P (y|x) is actually a dependent and multimodal distribution (Ling et al., 2013; Zen and Senior, 2014)1.”
Deep Learning in Mobile and Wireless Networking: A Survey (2018, background): “Powered by deep learning, the intelligent personal assistant Siri, developed by Apple, employs a deep mixture density networks [387] to fix typical robotic voice issues and synthesize more human-like voice [247].”
Practical Gauss-Newton Optimisation for Deep Learning (2017, methodology): “Our choice was motivated by recent interest in neural network mixture models (Eigen et al., 2013; Zen & Senior, 2014; van den Oord & Schrauwen, 2014; Shazeer et al., 2017); our mixture model is also appropriate for testing the performance of KFLR. Training was run for 40 , 000 updates for ADAM with…”
Espnet-TTS: Unified, Reproducible, and Integratable Open Source End-to-End Text-to-Speech Toolkit (2019, methodology): “Merlin is designed for DNN-based SPSS systems and it supports various types of neural networks including mixture density networks (MDN) [18] and recurrent neural networks (RNNs) [19], [20].”
Parameter Generation Algorithms for Text-To-Speech Synthesis with Recurrent Neural Networks (2018, methodology): “With the emergence of Deep Learning, the use of feed-forward Deep Neural Networks (DNN, [6]), Deep Mixture Density Networks (DMDN, [7]) and Long ShortTerm Memory (LSTM, [8]) RNN has naturally followed.”
Siri On-Device Deep Learning-Guided Unit Selection Text-to-Speech System (2017, methodology): “As a combination of a conventional neural network with a probability distribution, MDN provides a framework to model conditional density functions [10, 11].”
Statistical Parametric Speech Synthesis: A review (2015, methodology): “Mixture Density networks (MDNs) have been introduced for overcoming the limitations m DNN-based acoustic modelling for speech synthesis like the lack of variances and the unimodal nature of the objective function[7].”
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].”
Postfilters to Modify the Modulation Spectrum for Statistical Parametric Speech Synthesis (2016, methodology): “There are three main reasons causing the quality degradation [11]: parameterization errors in the speech analysis/synthesis stage [27], [28], [29], inaccurate modeling in the training stage [18], [30], and over-smoothness of the generated speech parameters in the synthesis stage [25], [26].”
Aligntts: Efficient Feed-Forward Text-to-Speech System Without Explicit Alignment (2020, methodology): “Since the training of Feed-Forward Transformer and the duration predictor requires the correct alignment between the text and mel-spectrum of training dataset, we design the mix density network [19, 20] to learn the alignment.”
A Review of Deep Learning Based Speech Synthesis (2019, methodology): “To address these problems, the authors in [44] proposed the parameter prediction method based on a deep mixture density network, which uses a mixture density output layer to predict the probability distribution of output features under given input features.”
Transformation Autoregressive Networks (2018, methodology): “The use of neural networks for density estimation dates back to early work by Bishop (1994) and has seen success in areas like speech (Zen & Senior, 2014; Uria, 2015), music (Boulanger-Lewandowski et al., 2012), etc. Typically such approaches use a network to learn the parameters of a parametric…”
On the training of DNN-based average voice model for speech synthesis (2016, result): “Recent studies have shown that SPSS has been considerably advanced by deep neural networks (DNN) [3], [5], [6], [7], [8], [9], [4].”
First Step Towards End-to-End Parametric TTS Synthesis: Generating Spectral Parameters with Neural Attention (2016, background): “A further extension is to improve the accuracy of acoustic feature distribution using mixture density networks (MDNs) [15, 18].”
Deep neural networks employing Multi-Task Learning and stacked bottleneck features for speech synthesis (2015, result): “These objective results are consistent with results in the literature [7, 25, 11].”
Edinburgh Research Explorer Modelling acoustic feature dependencies with artificial neural networks: Trajectory-RNADE (unknown): “Previous work has reported a beneﬁt from mixture models [8], although those models were ﬁtted with ten times more data than here.”