Unsupervised Vocal Tract Length Warped Posterior Features for Non-Parallel Voice Conversion

Abstract

In the non-parallel Voice Conversion (VC) with the Iterative combination of Nearest Neighbor search step and Conversion step Alignment (INCA) algorithm, the occurrence of one-tomany and many-to-one pairs in the training data will deteriorate the performance of the stand-alone VC system. The work on handling these pairs during the training is less explored. In this paper, we establish the relationship via intermediate speaker-independent posteriorgram representation, instead of directly mapping the source spectrum to the target spectrum. To that effect, a Deep Neural Network (DNN) is used to map the source spectrum to posteriorgram representation and another DNN is used to map this posteriorgram representation to the target speaker's spectrum. In this paper, we propose to use unsupervised Vocal Tract Length Normalization (VTLN)based warped Gaussian posteriorgram features as the speakerindependent representations. We performed experiments on a small subset of publicly available Voice Conversion Challenge (VCC) 2016 database. We obtain the lower Mel Cepstral Distortion (MCD) values with the proposed approach compared to the baseline as well as the supervised phonetic posteriorgram feature-based speaker-independent representations. Furthermore, subjective evaluation gave relative improvement of 13.3 % with the proposed approach in terms of Speaker Similarity (SS).

Cycle-consistent Adversarial Networks for Non-parallel Vocal Effort Based Speaking Style Conversion

“…The standard approach for non-parallel training in the domain of voice conversion is the INCA algorithm [16] (see Section 3.2.2) that iteratively finds alignments between individual frames in the source and target styles. Variants of the basic INCA may use several subsequent frames [17], dynamic features [18], or custom distance metrics [19] in the INCA alignment process. As a recent alternative to INCA, Cycle-consistent adversarial networks (CycleGANs, [20]) have shown promise in the domain of voice conversion.…”

Neural VTLN for Speaker Adaptation in TTS

“…Speaker dependent speech is then synthesized with an average model and a speaker specific warping over the whole phrase. In [9] two DNNs are trained to model the VTLN and reverse VTLN step for each speaker. As the normalized features are unknown, the authors propose an iterative unsupervised algorithm: 1.…”