Effect of data reduction on sequence-to-sequence neural TTS
Authors:
- Javier LatorreJLJavier Latorre
- Jakub LachowiczJLJakub Lachowicz
- Jaime Lorenzo-TruebaJLJaime Lorenzo-Trueba
- Thomas MerrittTMThomas Merritt
- Thomas DrugmanTDThomas Drugman
- Srikanth RonankiSRSrikanth Ronanki
- Klimkov ViacheslavKVKlimkov Viacheslav
Preprint
Abstract
Recent speech synthesis systems based on sampling from autoregressive neural networks models can generate speech almost undistinguishable from human recordings. However, these models require large amounts of data. This paper shows that the lack of data from one speaker can be compensated with data from other speakers. The naturalness of Tacotron2-like models trained on a blend of 5k utterances from 7 speakers is better than that of speaker dependent models trained on 15k utterances, but in terms of stability multi-speaker models are always more stable. We also demonstrate that models mixing only 1250 utterances from a target speaker with 5k utterances from another 6 speakers can produce significantly better quality than state-of-the-art DNN-guided unit selection systems trained on more than 10 times the data from the target speaker.Keywords
All Related Versions
This publication has 6 references indexed in Scilit:
- Towards Achieving Robust Universal Neural Vocoding2019
- Experiments with Training Corpora for Statistical Text-to-speech Systems.2018
- Investigating different representations for modeling and controlling multiple emotions in DNN-based speech synthesisSpeech Communication, 2018
- Natural TTS Synthesis by Conditioning Wavenet on MEL Spectrogram Predictions2018
- Tacotron: Towards End-to-End Speech Synthesis2017
- Analysis of Speaker Adaptation Algorithms for HMM-Based Speech Synthesis and a Constrained SMAPLR Adaptation AlgorithmIEEE Transactions on Audio, Speech, and Language Processing, 2009