Reformulating the HMM as a trajectory model by imposing explicit relationships between static and dynamic feature vector sequences

By Heiga Zen, Keiichi Tokuda, Tadashi Kitamura. 2004

187 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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

1. Junichi Yamagishi
2. Li Deng
3. Dong Yu
4. Heiga Zen
5. Yoshua Bengio
6. Tao Qin
7. Tie-Yan Liu
8. Zhenhua Ling

In Speech & Audio

1. Junichi Yamagishi
2. Heiga Zen
3. Tomoki Toda
4. Zhenhua Ling
5. Dong Yu
6. Keiichi Tokuda
7. Li Deng
8. Xin Eric Wang

In Signal Processing

1. Junichi Yamagishi
2. Heiga Zen
3. Dong Yu
4. Li Deng
5. Zhenhua Ling
6. Tomoki Toda
7. Simon King
8. Frank K. Soong

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).”
• Modeling Spectral Envelopes Using Restricted Boltzmann Machines and Deep Belief Networks for Statistical Parametric Speech Synthesis (2013, methodology): “trajectory model [9], improving the model training criterion by”
• Thousands of Voices for HMM-Based Speech Synthesis–Analysis and Application of TTS Systems Built on Various ASR Corpora (2010, methodology): “For computation of the mel-cepstral distance for the average voice itself, the random-sampling-based parameter generation algorithm [60] was used.”
• Voice Conversion Using Deep Neural Networks With Layer-Wise Generative Training (2014, background): “is a fixed matrix [33] and is the dimensionality of static feature vectors.”
• Statistical parametric speech synthesis with joint estimation of acoustic and excitation model parameters (2010, methodology): “Note that it is possible to model p (c | q, λ) and p (q | , λ) using existing acoustic models, such as HMM, HSMM, or trajectory HMM.”
• Autoregressive Models for Statistical Parametric Speech Synthesis (2013, methodology): “the probability of the set of all sequences of static features is not one [3].”
• A tutorial on task-parameterized movement learning and retrieval (2015, methodology): “The second technique is to encode both static and dynamic features in the mixture model as in trajectory-HMM [30, 95, 106, 89].”
• Statistical approach to vocal tract transfer function estimation based on factor analyzed trajectory HMM (2008, methodology): “The trajectory HMM [ 6 ] has a good property to model such a parameter sequence.”
• Integration of Spectral Feature Extraction and Modeling for HMM-Based Speech Synthesis (2014, methodology): “For the acoustic modeling, “trajectory HMM” [15]–[17] has been proposed as a generative model of static features considering the temporal continuity of feature sequences.”
• Statistical Parametric Speech Synthesis (2007, methodology): “Although it works better than the post-fi ltering technique (Toda et al., 2007), it still introduces some arti fi cial sounds into synthesized speech (Zen et al., 2006b).”
• Speech Synthesis Based on Hidden Markov Models (2013, methodology): “explicitly introduced the dynamic feature constraints into the training stage and reformulated the HMM with dynamic features as a trajectory model [21].”
• Deep neural networks employing Multi-Task Learning and stacked bottleneck features for speech synthesis (2015, methodology): “Significant efforts have been made to improve the acoustic models for SPSS, targeting the underlying model, the way the parameters are estimated during training, or the method for generating speech parameter trajectories when performing synthesis: minimum generation error training for HMM [4], globa”
• Edinburgh Research Explorer Modelling acoustic feature dependencies with artificial neural networks: Trajectory-RNADE (unknown): “Using these dynamical features, it is possible to construct a joint probabilistic model of the trajectory across time [5,6].”
• Support vector machines for noise robust ASR (2009, methodology): “Sampling from this distribution can be done efficiently using the approaches described in [19].”