Phoneme Recognizers followed by Language Modeling (PRLM) have consistently yielded top performance in language identification (LID) task. Parallel ordering of PRLMs (PPRLM) improves performance even more. Since tokenizer is the most important part of LID system the high quality phoneme recognizer is employed. Two different multilingual databases for training phoneme recognizers are compared and the amount of sufficient training data is studied. Reported results are on data from NIST 2003 LID evaluation. Our four PRLM systems have Equal Error Rate (EER) of 2.4 % on 12 languages task. This result compares favorably to the best known result from this task. 1.

Abstract. Language identification (LID) based on phono-tactic modeling is presented in this paper. Approaches using phoneme strings and strings of units automatically derived by an Ergodic HMM (EHMM) are compared. The phoneme recognizers were trained on 6 languages from OGI multi-language-corpus and Czech SpeechDat-E. The LID results are obtained on 4 languages. The results show superiority of Czech phoneme recognizer while used in LID and promising trends using the EHMMderived units. 1

This paper concentrates on PRLM (phoneme recognizer followed by language model) approach to language recognition. It elaborates on our prior work concerning the quality of phoneme recognition and amounts of training data for phoneme recognizer training. It reports improvements brought to our PRLM system by better phoneme recognition and Witten-Bell discounting in LM-modeling. The paper then concentrates on the use of phoneme lattices and anti-models. Training and scoring on phoneme lattices brought significant improvement in language recognition accuracy. The antimodels are simple, yet powerful technique to improve the discrimination between target and non-target languages. All results are reported on standard NIST 2003 data; comparison with other published results is favorable to our system. 1.

This paper reviews the current state of a Hungarian project which seeks to create a speech recognition system for the dictation of thyroid gland medical reports. First, we present the MRBA speech corpus that was assembled to support the training of general-purpose Hungarian speech recognition systems. Then we describe the processing of medical reports that were collected to help the creation of domain-specific language models. At the acoustic modelling level we experimented with two techniques – a conventional HMM one and an ANN-based solution – which are both briefly described in the paper. Finally, we present the language modelling methodology currently applied in the system, and round off with recognition results on test data taken from four speakers. The scores show that on a somewhat restricted sub-domain of the task we are able to produce word accuracies well over 95%. Povzetek: Prispevek predstavlja pregled trenutnega stanja madžarskega projekta, ki skuša vzpostaviti sistem razpoznavanja govora za narekovanje zdravniških izvidov na temo žleze ščitnice.