Analysis of Speaker Adaptation Algorithms for HMM-Based Speech Synthesis and a Constrained SMAPLR Adaptation Algorithm

By Junichi Yamagishi, Takao Kobayashi, Yuji Nakano, Katsumi Ogata, Juri Isogai. 2009

389 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

1. Junichi Yamagishi
2. Xu Tan
3. Tomoki Toda
4. Xin Eric Wang
5. Tao Qin
6. Tie-Yan Liu
7. Heiga Zen
8. Frank K. Soong

In Signal Processing

1. Junichi Yamagishi
2. Zhenhua Ling
3. Tomoki Toda
4. Xin Eric Wang
5. Heiga Zen
6. Xu Tan
7. Haizhou Li
8. Shinji Takaki

Citations in the corpus

• Neural Voice Cloning with a Few Samples (2018, methodology): “Speaker adaptation for voice cloning is in the same vein as these approaches, but differences arise when text-to-speech vs. speech-to-text are considered [Yamagishi et al., 2009].”
• Speech Synthesis Based on Hidden Markov Models (2013, methodology): “The major adaptation techniques used for HMM-based speech synthesis are similar to those of ASR and include maximum a posteriori (MAP) estimation [63], [64] and MLLR [26], [58].”
• Model architectures to extrapolate emotional expressions in DNN-based text-to-speech (2021, methodology): “The proposal included using a constrained structural maximum a posteriori linear regression (CSMAPLR) algorithm [16].”
• Spoofing and countermeasures for speaker verification: A survey (2015, background): “In contrast, state-of-theart HMM-based speech synthesisers (Zen et al., 2009; Yamagishi et al., 2009) can learn speech models from relatively little speaker-specific data by adapting background models derived from other speakers based on the standard model adaptation techniques drawn from speech rec”
• Statistical Parametric Speech Synthesis (2007, methodology): “Estimating linear-transformation matrices based on the MAP criterion (Yamagishi et al., 2009) and combining MAP estimation and MLLR have also been proposed (Ogata et al., 2006).”
• Speech-laughs: An HMM-based approach for amused speech synthesis (2015, methodology): “Previously trained models can hence be transformed into adapted models of the target voice through the Constrained Maximum Likelihood Linear Regression (CMLLR) method [13] (a description of the CMLLR adaptation algorithm can be found in [14]).”
• Investigating RNN-based speech enhancement methods for noise-robust Text-to-Speech (2016, positive): “We built an hidden Markov model (HMM)-based synthetic voice for the female and the male test data by adapting a previously trained model of an English female speaker clean data [33, 34].”
• Speech Enhancement for a Noise-Robust Text-to-Speech Synthesis System Using Deep Recurrent Neural Networks (2016, methodology): “We built an HMM-based synthetic voice for the female and the male data by adapting the model that was used previously used to obtained the phone-level alignments [32].”
• AdaDurIAN: Few-shot Adaptation for Neural Text-to-Speech with DurIAN (2020, background): “To scale the capacity for new speakers, we can adapt existing pre-trained multi-speaker sys-tem to generate new speakers’ voice, which is a well-studied subject of few-shot learning [13, 14] also known as speaker adaptation [15, 16].”
• Linear Networks Based Speaker Adaptation for Speech Synthesis (2018, methodology): “Besides, the Constrained Structural Maximum A Posterior Linear Regression (CSMAPLR) adaptation [7–9] simultaneously transforms the mean vectors and co-variance matrices of state output and state duration distributions of the HSMMs in speech synthesis for speaker adaptation.”
• Voice Morphing that improves TTS quality using an optimal dynamic frequency warping-and-weighting transform (2016, background): “Statistical speech models [6] can easily be adapted to a target speaker using speaker adaptation approaches developed for speech recognition [7].”
• Speech Enhancement of Noisy and Reverberant Speech for Text-to-Speech (2018, methodology): “It is possible to improve the quality of HMM-based TTS voices trained with noisy speech using adaptation techniques [2], [41], [42].”
• Unsupervised clustering of emotion and voice styles for expressive TTS (2012, methodology): “The models were 5 state left-to-right multi-space probability distribution hidden semi-Markov models (MSD-HSMM) [8].”
• The voice bank corpus: Design, collection and data analysis of a large regional accent speech database (2013, methodology): “With HMM-based speech synthesis, it is now possible to build a voice from a small amount of speech by adapting existing voice models pre-trained on many speakers [3].”
• Utterance Selection Techniques for TTS Systems Using Found Speech (2016, background): “Most of these previous techniques have concentrated on techniques for building average voices using a variety of speaker adaptation techniques[6],[7],[8].”