The 21st International Conference on Text, Speech and Dialogue (TSD 2018) took place the second week in September in Brno, Czech Republic. Of course we couldn’t miss it.
The central points of the conference were natural language processing (NLP), computational linguistics and speech technologies – and how these things can work together.
Current state of the art is represented by machine learning algorithms, so a large part of presentations was dedicated to their improvements and innovation. On the other hand, even less traditional methods of improving technologies were mentioned. One of the keynote speakers, Kenneth W. Church from the Chinese giant Baidu, brought up the often cited quotation “Every time I fire a linguist, the performance of our speech recognition system goes up.” The author of this quote, Fred Jelinek, one of the fathers of automatic speech recognition, was allegedly joking, but currently the trend indeed is to employ end-to-end systems which do not require any linguistic knowledge. We shall see what the future brings, but I’m not afraid for my job yet.
There were also interesting news about speech synthesis from our colleagues at University of West Bohemia in Pilsen. They tried the newest system WaveNet on Czech data, or – which was most interesting for me as a phonetician – began to include rules governing prosody of the synthesized sentences (melody, rhythm or loudness), so that they sound more natural.
We also saw some real-world applications. The Watson robot from IBM manages to diagnose cancer better than human doctors, and we could watch robot Leolani to exclaim “Humans are so confusing!” when given conflicted information by the researchers.
Speech recognition was one of the central topics of the conference, so we presented two posters prepared with our colleagues from the Technical University in Liberec. The first concerned robust speech recognition in noisy environment and the second dealt with errors in phonetic annotation and their influence on acoustic model training.
Lenka Weingartová, Linguist in NEWTON Technologies