In this paper, we present a novel technique for a non-parallel voice conversion (VC) with the use of cyclic variational autoencoder (CycleVAE)-based spectral modeling. In a variational autoencoder (VAE) framework, a latent space, usually with a Gaussian prior, is used to encode a set of input features. In a VAE-based VC, the encoded latent features are fed into a decoder, along with speaker-coding features, to generate estimated spectra with either the original speaker identity (reconstructed) or another speaker identity (converted). Due to the non-parallel modeling condition, the converted spectra can not be directly optimized, which heavily degrades the performance of a VAEbased VC. In this work, to overcome this problem, we propose to use CycleVAE-based spectral model that indirectly optimizes the conversion flow by recycling the converted features back into the system to obtain corresponding cyclic reconstructed spectra that can be directly optimized. The cyclic flow can be continued by using the cyclic reconstructed features as input for the next cycle. The experimental results demonstrate the effectiveness of the proposed CycleVAE-based VC, which yields higher accuracy of converted spectra, generates latent features with higher correlation degree, and significantly improves the quality and conversion accuracy of the converted speech.

In this paper, we propose a quasi-periodic neural network (QPNet) vocoder with a novel network architecture named pitchdependent dilated convolution (PDCNN) to improve the pitch controllability of WaveNet (WN) vocoder. The effectiveness of the WN vocoder to generate high-fidelity speech samples from given acoustic features has been proved recently. However, because of the fixed dilated convolution and generic network architecture, the WN vocoder hardly generates speech with given F0 values which are outside the range observed in training data. Consequently, the WN vocoder lacks the pitch controllability which is one of the essential capabilities of conventional vocoders. To address this limitation, we propose the PDCNN component which has the time-variant adaptive dilation size related to the given F0 values and a cascade network structure of the QPNet vocoder to generate quasi-periodic signals such as speech. Both objective and subjective tests are conducted, and the experimental results demonstrate the better pitch controllability of the QPNet vocoder compared to the same and double-size WN vocoders while attaining comparable speech qualities.

This paper presents the non-parallel voice conversion (VC) system developed at Nagoya University (NU) for the SPOKE task of the Voice Conversion Challenge 2018 (VCC2018). The goal of this task is to develop VC systems without the requirement of parallel training data. The key idea of our system is to use a text-to-speech (TTS) voice as a reference voice, making it possible to create two parallel training datasets between the source and TTS voices and between the TTS and target voices. Using these datasets, a cascade VC system is developed to convert the source voice into the target voice via the TTS voice as a reference. Furthermore, we also propose a system selection framework to avoid generating collapsed speech waveforms, which are often observed when using less accurately converted speech features in the WaveNet vocoder. In VCC2018, our system achieved the second best score for similarity (around 70%) and an above-average score for naturalness (a mean opinion score around 3.0) among the submitted systems.

No abstract

No abstract

This paper presents the NU (Nagoya University) voice conversion (VC) system for the HUB task of the Voice Conversion Challenge 2018 (VCC 2018). The design of the NU VC system can basically be separated into two modules consisting of a speech parameter conversion module and a waveformprocessing module. In the speech parameter conversion module, a deep learning framework is deployed to estimate the spectral parameters of a target speaker given those of a source speaker. Specifically, a deep neural network (DNN) and a deep mixture density network (DMDN) are used as the deep model structure. In the waveform-processing module, given the estimated spectral parameters and linearly transformed F0 parameters, the converted waveform is generated using a WaveNet-based vocoder system. To use the WaveNet-based vocoder, there are several generation flows based on an analysissynthesis framework to obtain the speech parameter set, on the basis of which a system selection process is performed to select the best one in an utterance-wise manner. The results of VCC 2018 ranked the NU VC system in second place with an overall mean opinion score (MOS) of 3.44 for speech quality and 85% accuracy for speaker similarity.

In this paper, we propose a technique to alleviate the quality degradation caused by collapsed speech segments sometimes generated by the WaveNet vocoder. The effectiveness of the WaveNet vocoder for generating natural speech from acoustic features has been proved in recent works. However, it sometimes generates very noisy speech with collapsed speech segments when only a limited amount of training data is available or significant acoustic mismatches exist between the training and testing data. Such a limitation on the corpus and limited ability of the model can easily occur in some speech generation applications, such as voice conversion and speech enhancement. To address this problem, we propose a technique to automatically detect collapsed speech segments. Moreover, to refine the detected segments, we also propose a waveform generation technique for WaveNet using a linear predictive coding constraint. Verification and subjective tests are conducted to investigate the effectiveness of the proposed techniques. The verification results indicate that the detection technique can detect most collapsed segments. The subjective evaluations of voice conversion demonstrate that the generation technique significantly improves the speech quality while maintaining the same speaker similarity. WaveNet vocoderWaveNet [8] is a deep autoregressive network capable of directly modeling a speech waveform sample-by-sample using the following conditional probability:

This paper presents a refinement framework of WaveNet vocoders for variational autoencoder (VAE) based voice conversion (VC), which reduces the quality distortion caused by the mismatch between the training data and testing data. Conventional WaveNet vocoders are trained with natural acoustic features but conditioned on the converted features in the conversion stage for VC, and such a mismatch often causes significant quality and similarity degradation. In this work, we take advantage of the particular structure of VAEs to refine WaveNet vocoders with the self-reconstructed features generated by VAE, which are of similar characteristics with the converted features while having the same temporal structure with the target natural features. We analyze these features and show that the self-reconstructed features are similar to the converted features. Objective and subjective experimental results demonstrate the effectiveness of our proposed framework.