Reformulating HMM as a trajectory model by imposing explicit relationships between static and dynamic features

2006

24 citations recorded by the source corpus.

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).”
• 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].”
• Statistical parametric speech synthesis: from HMM to LSTM-RNN (2015, unknown): “Trajectory HMM [18] • Derived from HMM by imposing dynamic feature constraints • Underlying generative model in HMM-based speech synthesis p(c | λ) = ∑ ∀q p(c | q, λ)P (q | λ) p(c | q, λ) = N (c; c̄q,Pq) where P−1 q = Rq = W >Σ−1 q W rq = W>Σ−1 q μq c̄q = Pqrq”
• Product of Experts for Statistical Parametric Speech Synthesis (2012, background): “Zen et al. [43] highlighted three major factors that degrade the quality of the synthesized speech; vocoding, accuracy of acoustic models (AMs), and over-smoothing.”
• Directly modeling speech waveforms by neural networks for statistical parametric speech synthesis (2015, methodology): “Typically, a minimum mean squared error (MMSE) or a maximum likelihood (ML) crite­ rion is used to estimate the model parameters [7] [8].”
• Continuous Stochastic Feature Mapping Based on Trajectory HMMs (2011, methodology): “Zen et al. reported that these inconsistencies degraded the quality of HMM-based speech synthesis [46].”
• A probabilistic interpretation for artificial neural network-based voice conversion (2015, background): “Moreover, it is noted that the probabilistic model for the proposed ANN-based VC may encounter an inconsistence between training and conversion when using a dynamic-feature constraint; such inconsistence can degrade the quality of converted speech [3, 4, 22].”