#marytts

Alexey Cherkes was the first speaker. He introduced new text-to-speech synthesis for Belarusian and Russian. It’s developed as a part of the project http://mbook.by/, a multimedia library for pupils. For the moment, the app is already available for Android. TTS is a nice feature of the app: finally, a pupil can listen school-books, not just read.

Alexey did a short overview of approaches to TTS, and told about modern tools and methods. Now we know that TTS feature is based on HTS; audience also enjoyed some online-demos of different synthesis systems. We were talking about the issue of a lack of linguistic resources for Belarusian and how to join our efforts to develop it. Also, we were told that prosody implementation is the next step to improve the quality of tts.

Demonstration of Belarusian tts for Android at #NLProc meetup pic.twitter.com/I4L4y4uf2Y

— nlproc.by (@nlprocby)

February 27, 2016

You can find the presentation here, but we can say that the value of it was actual live-demos.

The second speaker was Denis Postanogov, also very experienced NLProc-engineer. Denis with a team have been developing a full-featured commercial product for big business-players. But nobody was bored as it happens when people talk about business solutions, and actually people asked a lot of questions, Denis replied to all of them. We even had to extend the time of the meetup for an hour to give a chance to explain everything people asked about.

Sentiment analysis which is based on predictive QA in IHS Goldfire, discussing at #nlproc meetup in Minsk, @nlprocby pic.twitter.com/C0qdX0eG1t

— Yauhen (@YauhenMinsk)

27 February 2016

Here are just a couple of topics discussed: semantic relations, QA, machine translation, sentiment analysis, pdf parsing.

It’s hard to explain in this short post everything Denis was talking about, thus we recommend to look at the presentation and imagine the complexity of the product (attention: nice pics).

Traditionally, we were happy to say “Thank you!" to speakers by giving our nice T-shirts.

You can find more pictures here. Join us next time.

This part comprises the issues on a new voice creation, a toolset for a speech scientist, neural networks, evaluation approaches.

Enjoy it here.

Some mentioned entities:

HMM-based speech synthesis system (HTS), sptk, straight, eSpeak, indri, Taylor’s book.

Today we’d like to share the first part of the talk with our readers & listeners. 

With respect to an outline, Sebastien introduced himself at start, we continued by talking about the projects he was involved in, also the main focus of past and current research. Then we discussed the current state of MaryTTS, well-known text-to-speech synthesis system, and issues of a creating a new language.

Enjoy the talk.

Some links of the mentioned in the talk: Ingmar Steiner, DFKI 

The second (and last) part will be published very soon.

Topics we discussed in the second part include issues on new voice creation, a toolset for a speech scientist, neural networks, evaluation approaches.