Speaker-independent raw waveform model for glottal excitation

By Lauri Juvela, Vassilis Tsiaras, Bajibabu Bollepalli, Manu Airaksinen, Junichi Yamagishi, Paavo Alku. 2018

39 citations recorded by the source corpus.

Recent speech technology research has seen a growing interest in using WaveNets as statistical vocoders, i.e., generating speech waveforms from acoustic features. These models have been shown to improve the generated speech quality over classical vocoders in many tasks, such as text-to-speech synthesis and voice conversion. Furthermore, conditioning WaveNets with acoustic features allows sharing the waveform generator model across multiple speakers without additional speaker codes. However, multi-speaker WaveNet models require large amounts of training data and computation to cover the entire acoustic space. This paper proposes leveraging the source-filter model of speech production to more effectively train a speaker-independent waveform generator with limited resources. We present a multi-speaker 'GlotNet' vocoder, which utilizes a WaveNet to generate glottal excitation waveforms, which are then used to excite the corresponding vocal tract filter to produce speech. Listening tests show that the proposed model performs favourably to a direct WaveNet vocoder trained with the same model architecture and data.

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

In Signal Processing

Citations in the corpus

LPCNET: Improving Neural Speech Synthesis through Linear Prediction (2018, background): “arXiv:1810.11846v2 [eess.AS] 19 Feb 2019 3.2. Pre-emphasis and Quantization ! " ! # $% approaches [12]. Fig. 1. Overview of the LPCNet algorithm. The left part of the network (yellow) is computed once per frame and its result is held constant throughout the frame for t”
A Real-Time Wideband Neural Vocoder at 1.6 kb/s Using LPCNet (2019, methodology): “Other models use linear prediction to remove the burden of spectral envelope modeling from the neural synthesis network [12, 13, 14].”
FeatherWave: An efficient high-fidelity neural vocoder with multi-band linear prediction (2020, methodology): “Instead of openloop filtering approaches [16], LPCNet preforms as a closedloop synthesis of predicting sample xt by conditioning on the previously sampled excitation et−1 and current prediction pt, which can improve the quality of generated speech.”
Improving FFTNet Vocoder with Noise Shaping and Subband Approaches (2018, methodology): “…network structure modiﬁcations such as the gated activation units used in WaveNet [17] and signal processing techniques such as subband methods with multiband input combined with noise shaping will be investigated at higher sampling frequencies, such as 24 [18,25,34] and 48 kHz [24, 47].”
GELP: GAN-Excited Liner Prediction for Speech Synthesis from Mel-spectrogram (2019, methodology): “Mean-while, the source-ﬁlter model has also been used with linear predictive (LP) ﬁlters to build neural models for the glottal excitation [16, 17] or the LP residual signal [18, 19].”
ExcitNet Vocoder: A Neural Excitation Model for Parametric Speech Synthesis Systems (2018, methodology): “There have been prior works in using the spectral filters in the WaveNet applications [18, 25].”
GlotNet—A Raw Waveform Model for the Glottal Excitation in Statistical Parametric Speech Synthesis (2019, methodology): “Furthermore, we feel that the WaveNet vocoder was under-performing in [34], partially due to the high acoustic”
Neural Text-to-Speech with a Modeling-by-Generation Excitation Vocoder (2020, background): “More recently, neural excitation vocoders such as GlotNet, ExcitNet, LP-WaveNet and LPC-Net [6–10] have exploited the advantages of linear prediction (LP)-based parametric vocoders.”
GELP: GAN-Excited Linear Prediction for Speech Synthesis from Mel-spectrogram (2019, methodology): “Meanwhile, the source-filter model has also been used with linear predictive (LP) filters to build neural models for the glottal excitation [16, 17] or the LP residual signal [18, 19].”