Statistical Parametric Speech Synthesis

By Heiga Zen, Keiichi Tokuda, Alan W. Black. 2007

1,584 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Natural Language Generation, Speech & Audio

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

In Natural Language Generation

In Speech & Audio

Citations in the corpus

Natural TTS Synthesis by Conditioning Wavenet on MEL Spectrogram Predictions (2017, background): “Statistical parametric speech synthesis [4, 5, 6, 7], which directly generates smooth trajectories of speech features to be synthesized by a vocoder, followed, solving many of the issues that concatenative synthesis had with boundary artifacts.”
Tacotron: Towards End-to-End Speech Synthesis (2017, methodology): “For example, it is common for statistical parametric TTS to have a text frontend extracting various linguistic features, a duration model, an acoustic feature prediction model and a complex signal-processing-based vocoder (Zen et al., 2009; Agiomyrgiannakis, 2015).”
Neural Speech Synthesis with Transformer Network (2018, methodology): “Parametric TTS systems (Tokuda et al. 2000; Zen, Tokuda, and Black 2009; Ze, Senior, and Schuster 2013; Tokuda et al. 2013) convert speech waves into spectrograms, and acoustic parameters, such as fundamental frequency and duration, are used to synthesize new audio results.”
ELLA-V: Stable Neural Codec Language Modeling with Alignment-guided Sequence Reordering (2024, methodology): “With the advancement of modern neural networks, some work initially replaced HMMs with recurrent neural networks (RNNs) but still followed the SPSS paradigm (Fan et al., 2014; Zen and Sak, 2015; Valentini-Botinhao et al., 2016).”
VQTTS: High-Fidelity Text-to-Speech Synthesis with Self-Supervised VQ Acoustic Feature (2022, methodology): “Compared with traditional statistical parametric speech synthesis [1], neural TTS model [2, 3] based on deep neural network shows a better performance.”
A Survey on Neural Speech Synthesis (2021, background): “While there are multiple survey papers on statistical parametric speech synthesis [27, 418, 351, 415] and neural TTS [325, 220, 300, 242, 115, 254, 236], a comprehensive survey on the basics and the recent developments of neural TTS is still necessary since the topics in this area are diverse and ev”
LibriTTS: A Corpus Derived from LibriSpeech for Text-to-Speech (2019, methodology): “The CMU ARCTIC corpus [19] has been used for many years in statistical parametric speech synthesis research [28].”
A Spectral Energy Distance for Parallel Speech Synthesis (2020, neutral): “Text-to-speech synthesis (TTS) has seen great advances in recent years, with neural network-based methods now significantly outperforming traditional concatenative and statistical parametric approaches [39, 35].”
Adversarial Audio Synthesis (2018, background): “Parametric systems map text to salient parameters of speech, which are then synthesized by a vocoder (Dudley, 1939); see (Zen et al., 2009) for a comprehensive review.”
Hierarchical Generative Modeling for Controllable Speech Synthesis (2018, background): “The second Global Style Token (GST) model constrains a reference embedding to be a weighted combination of a fixed set of learned vectors, while the third further restricts the weights to be one-hot, and is built on a conventional parametric speech synthesizer (Zen et al., 2009).”
Fast, Compact, and High Quality LSTM-RNN Based Statistical Parametric Speech Synthesizers for Mobile Devices (2016, methodology): “t;0:5 and c>1. Note that if = 0 and = I, the -contaminated Gaussian loss function is equivalent to the squared loss function. Figure 4 illustrates -contaminated Gaussian distribution (= [0], = [1], c= 10 and = 0:1). It can be seen from the ﬁgure that the -contaminated Gaussian distribution has he”
Voice Synthesis for in-the-Wild Speakers via a Phonological Loop (2017, methodology): “While HMM-based methods [24] require careful collection of the samples, concatenative methods are somewhat less restrictive but still require tens of minutes of clean and well transcribed samples from the target voice.”