Speech Synthesis also called TTS (Text-to-speech) virtually deals with computer-generated simulation of natural human speech. It is used to automatically read texts, in automotive satellite navigation devices but also in language learning. Lingea has been engaging in this field for almost 2 years now, especially in relation to foreign language learning.

Currently, we have available a set of 6 different voices for English, German and Russian. In cooperation with the University of West Bohemia, we have managed to improve the quality of the audio output to such a degree that it can virtually replace real human voices recorded in a studio. The voices are optimized for the purposes of language learning but they are also fully exploitable for other types of applications.

The current TTS version is based on the concatenative synthesis approach which still delivers more natural-sounding results and moreover allows post-editing to achieve perfect output. However, simultaneously experiments are being conducted using neural networks (already successfully used in Machine Translation). These should bring improved possibilities of output modelling as well as the creation of less memory-intensive models.

  • High quality output usable for foreign language learning
  • Possible modifications of audio output by editing the corresponding IPA transcriptions
  • Extensive first-rate pronunciation dictionaries for 30 languages