Neural Speech Synthesis for Estonian

TL;DR

This paper presents new neural speech synthesis models for Estonian, supported by a large, openly available dataset from multiple speakers, and compares their performance to existing solutions.

Contribution

It introduces a sizable, openly released Estonian speech dataset and open-source neural synthesis models, along with comprehensive evaluation results.

Findings

Openly released 92.4 hours of speech data from 6 speakers

Neural models outperform HMM-based and Google TTS in evaluations

Published open-source software and models for Estonian speech synthesis

Abstract

This technical report describes the results of a collaboration between the NLP research group at the University of Tartu and the Institute of Estonian Language on improving neural speech synthesis for Estonian. The report (written in Estonian) describes the project results, the summary of which is: (1) Speech synthesis data from 6 speakers for a total of 92.4 hours is collected and openly released (CC-BY-4.0). Data available at https://konekorpus.tartunlp.ai and https://www.eki.ee/litsents/. (2) software and models for neural speech synthesis is released open-source (MIT license). Available at https://koodivaramu.eesti.ee/tartunlp/text-to-speech . (3) We ran evaluations of the new models and compared them to other existing solutions (HMM-based HTS models from EKI, http://www.eki.ee/heli/, and Google's speech synthesis for Estonian, accessed via https://translate.google.com). Evaluation includes voice acceptability MOS scores for sentence-level and longer excerpts, detailed error analysis and evaluation of the pre-processing module.