Multi-Space Probability Distribution HMM

By Keiichi Tokuda, Takashi Masuko, Noboru Miyazaki, Takao Kobayashi. 2002

295 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/33ae9170c2c4e5bf0fbc13085f1002ae52a298e1

How researchers see this paper

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

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

Citations in the corpus

• Deep mixture density networks for acoustic modeling in statistical parametric speech synthesis (2014, methodology): “To model logF0 sequences consisting of voiced and unvoiced observations, a multi-space probability distribution (MSD) was used [27].”
• Statistical parametric speech synthesis using deep neural networks (2013, methodology): “tions, a multi-space probability distribution (MSD) was used [36].”
• Current trends in multilingual speech processing (2011, background): “…and • Models: normally differ from standard HMM by including explicit duration modelling (hidden semi-Markov model – HSMM (Zen et al 2007)) and provide appropriate distributions for modelling of discontinuous features such as pitch (multi-space probability distribution – MSD (Tokuda et al 2002)).”
• A Vector Quantized Variational Autoencoder (VQ-VAE) Autoregressive Neural $F_0$ Model for Statistical Parametric Speech Synthesis (2020, methodology): “Specifically, multi-space probability distribution HMMs directly model the rawF0 contours that may contain unvoiced frames [15], whereas HMM-based continuous F0 models use interpolated F0 contours and the voicing status as the target [16].”
• Statistical parametric speech synthesis: from HMM to LSTM-RNN (2015, methodology): “Experimental setup Database US English female speaker Training / test data 33000 & 173 sentences Sampling rate 16 kHz Analysis window 25-ms width / 5-ms shift Linguistic 11 categorical features features 25 numeric features Acoustic 0–39 mel-cepstrum features logF0, 5-band aperiodicity, ∆,∆2 HMM 5-st”
• Analysis of Speaker Adaptation Algorithms for HMM-Based Speech Synthesis and a Constrained SMAPLR Adaptation Algorithm (2009, methodology): “In the HMM-based CMLLR adaptation [37], [38], mean vectors and covariance matrices of the state output pdfs are transformed simultaneously using the same matrix (Fig.”
• Speech Synthesis Based on Hidden Markov Models (2013, unknown): “probability distributions [50] for modeling them.”
• Autoregressive Models for Statistical Parametric Speech Synthesis (2013, background): “MCD provides useful information about the accuracy of the mean trajectories independent of the trajectory covariances.”
• Statistical Parametric Speech Synthesis (2007, methodology): “Although several methods have been investigated for modeling sequences (Freij and Fallside, 1988; Jensen et al., 1994; Ross and Ostendorf, 1994), the HMM-based speech synthesis sys-tem adopts multi-space probability distributions (Tokuda et al., 2002a) as their stream-output distributions.”
• Robust Speaker-Adaptive HMM-Based Text-to-Speech Synthesis (2009, methodology): “2016394 well as the spectral parameters, the multispace probability distribution (MSD) HMM [10] was developed.”
• Multi-distribution deep belief network for speech synthesis (2013, methodology): “The core underpinning techniques include: (1) the adoption of multi-space distribution HMMs in estimating the statistical behavior of speech parameters [2]; (2) the parameter generation algorithm which uses dynamic features to smooth the originally piece-wise constant speech parameters drawn from…”