The advancement of automatic speech recognition (ASR) has been largely enhanced by extensive datasets in high-resource languages, while languages such as Hungarian remain underrepresented due to limited spontaneous and conversational corpora. To address this gap, we introduce two new datasets -- BEA-Large and BEA-Dialogue -- constructed from the previously unprocessed portions of the Hungarian speech corpus named BEA. BEA-Large extends BEA-Base with 255 hours of spontaneous speech from 433 speakers, enriched with detailed segment-level metadata. BEA-Dialogue, comprising 85 hours of spontaneous conversations, is a Hungarian speech corpus featuring natural dialogues partitioned into speaker-independent subsets, supporting research in conversational ASR and speaker diarization. We establish reproducible baselines on these datasets using publicly available ASR models, with the fine-tuned Fast Conformer model achieving word error rates as low as 14.18% on spontaneous and 4.8% on repeated speech. Diarization experiments yield diarization error rates between 12.46% and 17.40%, providing reference points for future improvements. The results highlight the persistent difficulty of conversational ASR, particularly due to disfluencies, overlaps, and informal speech patterns. By releasing these datasets and baselines, we aim to advance Hungarian speech technology and offer a methodological framework for developing spontaneous and conversational benchmarks in other languages.

Even though various speech data sets are available in Hungarian, there is a lack of a general overview about their types and sizes. To fill in this gap, we provide a survey of available data sets in spoken Hungarian in five categories (e.g., monolingual, Hungarian part of multilingual, pathological, child-related and dialectal collections). In total, the estimated size of available data is about 2800 hours (across 7500 speakers) and it represents a rich spoken language diversity. However, the distribution of the data and its alignment to real-life (e.g. speech recognition) tasks is far from optimal indicating the need for additional larger-scale natural language speech data sets. Our survey presents an overview of available data sets for Hungarian explaining their strengths and weaknesses which is useful for researchers working on Hungarian across disciplines. In addition, our survey serves as a starting point towards a unified foundational speech model specific to Hungarian.

Hungarian is spoken by 15 million people, still, easily accessible Automatic Speech Recognition (ASR) benchmark datasets – especially for spontaneous speech – have been practically unavailable. In this paper, we introduce BEA-Base, a subset of the BEA spoken Hungarian database comprising mostly spontaneous speech of 140 speakers. It is built specifically to assess ASR, primarily for conversational AI applications. After defining the speech recognition subsets and task, several baselines – including classic HMM-DNN hybrid and end-to-end approaches augmented by cross-language transfer learning – are developed using open-source toolkits. The best results obtained are based on multilingual self-supervised pretraining, achieving a 45% recognition error rate reduction as compared to the classical approach – without the application of an external language model or additional supervised data. The results show the feasibility of using BEA-Base for training and evaluation of Hungarian speech recognition systems.

Freely available audiobooks are a rich resource of expressive speech recordings that can be used for the purposes of speech synthesis. Natural sounding, expressive synthetic voices have previously been built from audiobooks that contained large amounts of highly expressive speech recorded from a professionally trained speaker. The majority of freely available audiobooks, however, are read by amateur speakers, are shorter and contain less expressive (less emphatic, less emotional, etc.) speech both in terms of quality and quantity. Synthesizing expressive speech from a typical online audiobook therefore poses many challenges. In this work we address these challenges by applying a method consisting of minimally supervised techniques to align the text with the recorded speech, select groups of expressive speech segments and build expressive voices for hidden Markov-model based synthesis using speaker adaptation. Subjective listening tests have shown that the expressive synthetic speech generated with this method is often able to produce utterances suited to an emotional message. We used a restricted amount of speech data in our experiment, in order to show that the method is generally applicable to most typical audiobooks widely available online.

Various morphological and acoustic modeling techniques are evaluated on a less resourced, spontaneous Hungarian large-vocabulary continuous speech recognition (LVCSR) task. Among morphologically rich languages, Hungarian is known for its agglutinative, inflective nature that increases the data sparseness caused by a relatively small training database. Although Hungarian spelling is considered as simple phonological, a large part of the corpus is covered by words pronounced in multiple, phonemically different ways. Data-driven and language specific knowledge supported vocabulary decomposition methods are investigated in combination with phoneme- and grapheme-based acoustic modeling techniques on the given task. Word baseline and morph-based advanced baseline results are significantly outperformed by using both statistical and grammatical vocabulary decomposition methods. Although the discussed morph-based techniques recognize a significant amount of out of vocabulary words, the improvements are due not to this fact but to the reduction of insertion errors. Applying grapheme-based acoustic models instead of phoneme-based models causes no severe recognition performance deteriorations. Moreover, a fully data-driven acoustic modeling technique along with a statistical morphological modeling approach provides the best performance on the most difficult test set. The overall best speech recognition performance is obtained by using a novel word to morph decomposition technique that combines grammatical and unsupervised statistical segmentation algorithms. The improvement achieved by the proposed technique is stable across acoustic modeling approaches and larger with speaker adaptation.

A coupled acoustic- and language-modeling approach is presented for the recognition of spontaneous speech primarily in agglutinative languages. The effectiveness of the approach in large vocabulary spontaneous speech recognition is demonstrated on the Hungarian MALACH corpus. The derivation of morphs from word forms is based on a statistical morphological segmentation tool while the mapping of morphs into graphemes is obtained trivially by splitting each morph into individual letters. Using morphs instead of words in language modeling gives significant WER reductions in case of both phoneme- and grapheme-based acoustic modeling. The improvements are larger after speaker adaptation of the acoustic models. In conclusion, morphophonemic and the proposed morpho-graphemic ASR approaches yield the same best WERs, which are significantly lower than the word-based baselines but essentially without language dependent rules or pronunciation dictionaries in the latter case. Index Terms: spontaneous speech recognition, morphology.

Medicine Line (MLN) is an automatic telephone information system operating in Hungary since December 2006. It is primarily intended for visually handicapped persons and elderly people. In Hungary the National Institute of Pharmacy (NIP) coordinates the approval of new drugs and also their Patient Information Leaflets (PIL). Medicine Line reads this textual information chapter by chapter to the citizens. The number of different medicines used in Hungary is about 5000. New drugs come into practice regularly, and some are withdrawn after a certain time. The MLN system ensures 24 hour access to the information. The spoken dialogue input is processed by a specialized ASR module (the caller tells the name of the drug, the chapter title etc.). The output is given by a TTS synthesizer specialized to read drug names and medical Latin words correctly. The user can control the system by DTMF buttons too. In this article we will focus on features of speech based components. Index Terms: medical TTS, ASR for drug names, combination of TTS and ASR, pronunciation of drugs, Patient

In this paper a variety of front-end configurations are evaluated on Hungarian telephone speech databases. Our aim was to measure directly the efficiency of the front-ends on real noisy and normal speech data. As a baseline the ETSI ADSR standard front-end is used. Some simplification on the standard is introduced resulting in better performance on our databases than the original front-end in terms of both speed and recognition rate. Besides, another recently proposed feature extraction approach is also investigated. Finally the effect of the novel voice activity detection approach is evaluated. The best front-end configuration augmented with this voice activity detector outperformed significantly the baseline in each recognition test and by 24,7% relative in average.

A novel noise robust voice activity detection approach is introduced. The novelty of the method that it uses noise suppressed spectrum of the input signal for spectral entropy calculation. As a result excellent end-pointing performance is observed based on predefined global entropy threshold and time constraints. The effect of frame dropping controlled by the proposed algorithm was investigated on the accuracy of automatic speech recognition. The experiments were performed on Hungarian publicly available noisy and normal telephony speech databases. The relative improvement due to dropping of non-speech frames was positive in all test configurations with a maximum of 29,5%. Besides, in average more than 50% of the frames were dropped.

Abstract In this article we present a voice enabled call center whichintegrates our basic and applied research results on Hungarianspeech recognition. Telephone interfaces, data storage and re-trieval modules, and an intelligent dialog descriptor and man-ager module are also parts of the system. To evaluate the ef-ficiency of the recognition and the dialog, a voice enabled callcenter was implemented and tested under real life conditions.This article describes the main modules of the system and com-pares the result of the field tests with that of the laboratory test-ing. 1. Introduction Hungarian speech research has a long tradition at our Labora-tory, there exist some research systems both for speech synthe-sis and for ASR (automatic speech recognition). Our previousASR research results were focused on special problems, likeacoustic model training, language modeling [1], and pronuncia-tion modeling [3]. Recently, large Hungarian speech databaseshave become available, recorded over mobile and fixed tele-phone channels, which encouraged the implementation of di-alog systems that can be used for various services in the field oftelephony. On one hand we had to integrate our previous basicand applied research results while, on the other hand, applyingreal voice driven dialogs initiated a new research area for Hun-garian.In the following section we provide an overview of thecall center realized, going into details about the main compo-nents, especially about the speech recognition engine. Then wedemonstrate our experimental results of the system and comparethem with laboratory test results.

In recent speech recognition systems the base unit of recognition is generally the speech sound. To each speech sound an acoustic model is associated, whose parameters are estimated by statistical methods. The proper training data fundamentally determine the efficiency of the recognizer. Present day technology and computational capacity allow speech recognition systems to operate with large dictionaries and complex language models, but the quality of the basic pattern matching units has large influence on the reliability of the system. In our experiments presented here we investigated the effects of different training methods to the recognition accuracy; namely, the effect of increasing the number of speakers and the number of mixtures were examined in the case of pronunciation modeling and context independent models.

In Hungarian, as more or less in many other languages, a large percent of words and phrases can be pronounced in several, different, but correct ways. Introducing pronunciation alternatives for individual vocabulary elements may improve the efficiency of the recognition. But in connected word recognition tasks the modeling of inter-word phonetic changes has a greater significance. In this paper we introduce a rule-based method for the automatic generation of pronunciation alternatives used first for isolated words and later the method is extended to handle cross-word phonological changes in recognition networks, applying a special approach applicable for the Hungarian language. To evaluate the method it is tested in connected number recognition tests.