Residual-Based Excitation with Continuous F0 Modeling in HMM-Based Speech Synthesis

Csapó, Tamás Gábor; Németh, Géza; Cerňak, Miloš

doi:10.1007/978-3-319-25789-1_4

Cited by 12 publications

(16 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…During the synthesis phase, voiced excitation is composed of residual excitation frames overlap-added pitch synchronously, depending on the continuous F0 [28,29,30]. After that, this voiced excitation is lowpass filtered frame by frame at the frequency given by the MVF parameter.…”

Section: Continuous Vocodermentioning

confidence: 99%

“…The RMSE of the Maximum Voiced Frequency prediction is in the range of 654-1177 Hz, indicating that for some speakers the MVF can be estimated with lower error, while for Female #1 this task was more difficult. Comparing the UTI-to-MVF prediction results with text-to-MVF prediction (within HMM-based speech synthesis [28,29]), the latter seems to be a simpler task. Predicting voicing related parameters from articulatory data can be done only through indirect relationships, as pointed out in Section 1.…”

Section: Objective Evaluationmentioning

confidence: 99%

See 1 more Smart Citation

Ultrasound-Based Silent Speech Interface Built on a Continuous Vocoder

et al. 2019

Self Cite

View full text Add to dashboard Cite

Recently it was shown that within the Silent Speech Interface (SSI) field, the prediction of F0 is possible from Ultrasound Tongue Images (UTI) as the articulatory input, using Deep Neural Networks for articulatory-to-acoustic mapping. Moreover, text-to-speech synthesizers were shown to produce higher quality speech when using a continuous pitch estimate, which takes non-zero pitch values even when voicing is not present. Therefore, in this paper on UTI-based SSI, we use a simple continuous F0 tracker which does not apply a strict voiced / unvoiced decision. Continuous vocoder parameters (ContF0, Maximum Voiced Frequency and Mel-Generalized Cepstrum) are predicted using a convolutional neural network, with UTI as input. The results demonstrate that during the articulatory-toacoustic mapping experiments, the continuous F0 is predicted with lower error, and the continuous vocoder produces slightly more natural synthesized speech than the baseline vocoder using standard discontinuous F0.

show abstract

Section: Continuous Vocodermentioning

confidence: 99%

Section: Objective Evaluationmentioning

confidence: 99%

Ultrasound-Based Silent Speech Interface Built on a Continuous Vocoder

et al. 2019

Self Cite

View full text Add to dashboard Cite

show abstract

“…20 sentences from each speaker were chosen randomly to be analyzed and synthesized with the baseline and proposed vocoders. These 60 utterances were subsequently down-sampled by a factor of 2 in order to reduce its sampling rate from 32 kHz to 16 kHz, as this is a more typical use in the early baseline vocoder [13].…”

Section: Datasetsmentioning

confidence: 99%

“…Tóth and Csapó [12] have shown that continuous F0 contour can be better approximated with HMM and deep neural network (DNN) than traditional discontinuous F0. In [13], an excitation model has been proposed which combines continuous F0 modeling with Maximum Voiced Frequency (MVF). This model has been shown to produce more natural synthesized speech for voiced sounds than traditional vocoders based on standard pitch tracking, whereas it was also found that there is a room for improvement in modeling unvoiced sounds.…”

Section: Introductionmentioning

confidence: 99%

“…In our recent work in SPSS, we proposed a computationally feasible residual-based vocoder [13] using a continuous F0 model [9] and maximum voiced frequency (MVF) [30]. In this method, the voiced excitation consisting of pitch synchronous residual frames is low-pass filtered while the unvoiced part is high-pass filtered according to the MVF contour as a cutoff frequency.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Adaptive Refinements of Pitch Tracking and HNR Estimation within a Vocoder for Statistical Parametric Speech Synthesis

2019

Self Cite

View full text Add to dashboard Cite

Recent studies in text-to-speech synthesis have shown the benefit of using a continuous pitch estimate; one that interpolates fundamental frequency (F0) even when voicing is not present. However, continuous F0 is still sensitive to additive noise in speech signals and suffers from short-term errors (when it changes rather quickly over time). To alleviate these issues, three adaptive techniques have been developed in this article for achieving a robust and accurate F0: (1) we weight the pitch estimates with state noise covariance using adaptive Kalman-filter framework, (2) we iteratively apply a time axis warping on the input frame signal, (3) we optimize all F0 candidates using an instantaneous-frequency-based approach. Additionally, the second goal of this study is to introduce an extension of a novel continuous-based speech synthesis system (i.e., in which all parameters are continuous). We propose adding a new excitation parameter named Harmonic-to-Noise Ratio (HNR) to the voiced and unvoiced components to indicate the degree of voicing in the excitation and to reduce the influence of buzziness caused by the vocoder. Results based on objective and perceptual tests demonstrate that the voice built with the proposed framework gives state-of-the-art speech synthesis performance while outperforming the previous baseline.

show abstract