Edinburgh Research Explorer Modelling acoustic feature dependencies with artificial neural networks: Trajectory-RNADE

By Benigno Uria, Iain Murray, Steve Renals, Cassia Valentini-Botinhao, John Scott Bridle

Subfields: Natural Language Processing

How researchers see this paper

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

In Natural Language Processing

1. Yoshua Bengio
2. Kyunghyun Cho
3. Oriol Vinyals
4. Yoav Goldberg
5. Quoc V. Le
6. Li Deng
7. Graham Neubig
8. Dong Yu

Citations in the corpus

• WaveNet: A Generative Model for Raw Audio (2016, background): “…developing high-quality vocoders (Kawahara et al., 1999; Agiomyrgiannakis, 2015; Morise et al., 2016), improving the accuracy of generative models (Zen et al., 2007; 2013; Fan et al., 2014; Uria et al., 2015), and compensating the oversmoothing effect (Toda & Tokuda, 2007; Takamichi et al., 2016).”
• Masked Autoregressive Flow for Density Estimation (2017, background): “Combining the flexibility and learning capacity of neural networks with prior knowledge about the structure of data to be modelled has led to impressive results in modelling natural images [4, 30, 37, 38] and audio data [34, 36].”
• Connectionist multivariate density-estimation and its application to speech synthesis (2016, background): “This chapter is an extended version of the article “Modelling acoustic feature dependencies with artificial neural networks: Trajectory-RNADE” (Uria et al., 2015) published in ICASSP 2015.”
• Investigating gated recurrent neural networks for speech synthesis (2016, unknown): “In [10] and [11], mixture density networks (MDNs) and realvalued neural autoregressive density estimators (RNADEs) were proposed, respectively, to predict acoustic feature distributions given input linguistic features.”
• 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”