Alex Dialogue Systems Framework - release 2013
Pre-releaseThis is the Alex Dialogue Systems Framework - release 2013.
This framework is being developed by the dialogue systems group at UFAL - http://ufal.mff.cuni.cz/ - the Institute of Formal and Applied Linguistics, Faculty of Mathematics and Physics, Charles University in Prague, Czech Republic. The purpose of this work is to facilitate research into and development of spoken dialogue systems.
The main goals are: to provide baseline components need for a building spoken dialogue systems (SDSs), to provide example implementations of SDSs for several domains, to provide tools for processing dialogue system interactions logs, e.g. for audio transcription, semantic annotation, or SDSs evaluation.
Implemented features: VOIP using PJSIP 2.1 with some modifications, ASR using OpenJulius, GoogleASR or KALDI, VAD using Gaussian Mixture Models or Feed-Forward Neural Networks, SLU using a set of logistic regression classifiers for detecting dialogue acts, DM using probabilistic discriminative dialogue state tracking and handcrafted policies, NLG using template based generation possibly with efficient inflection into the correct surface form for morphologically rich languages, TTS using flite, VoiceRSS and SpeechTech, evaluation of dialogue systems using Amazon Mechanical Turk crowdsourcing platform, building acoustic models using the HTK and KALDI toolkits.
This work was partly funded by the Ministry of Education, Youth and Sports of the Czech Republic under the grant agreement LK11221 and core research funding of Charles University in Prague.