Is Spoken Hungarian Low-resource?: A Quantitative Survey of Hungarian Speech Data Sets
Authors: Péter Mihajlik, Katalin Mády, Anna Kohári, Fruzsina Sára Vargha, Gábor Kiss, Tekla Etelka Gráczi, A. Seza Doğruöz
Organizations: HUN-REN Hungarian Research Centre for Linguistics · Budapest University of Technology and Economics, Departure of Telecommunications and Media Informatics · LT3, IDLab, Universiteit Gent
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.
Figures & tables
Name
[hours]
Speakers
Key Features
BUSZI
~600
250
Sociolinguistic Interviews recorded with magnetofon. Kontra and Váradi (1997)
Oasis Numbers
3
26
Domain-specific corpus of numbers, 5857 recorded words. Kocsor et al. (2000)
MTBA
~5
500
Mostly read telephone speech. Vicsi et al. (2002)
MRBA
~6
332
Phonetically balanced read text. Vicsi et al. (2004)
BME Broadcast
3
~10
TV broadcast news. Teleki et al. (2005)
Szeged Broadcast
28
n.a.
News broadcast from 8 tv channels. Gosztolya and Tóth (2010)
Table 1: Monolingual Hungarian Data Sets
Name
[hours]
Speakers
Key Features
BABEL
1
72
Clear speech, read text. Vicsi and Vig (1998)
SpeechDat(E)
65
1000
Read telephone speech, ELRA. Pollák et al. (2000)
SPEECON-1
~200
555
Read speech recorded through 4 microphone channels, ELRA. Speecon Consortium (2000)
CSLU: 22 Languages
~4
300
Prompted telephone speech, LDC. Lander (2005)
CSS10
10
1
Free audiobook. Park and Mulc (2019)
CommonVoice
151
1603
Read sentences, online collection. Ardila et al. (2019)
Table 2: Hungarian Data Sets as Parts of Multilingual Collections
Name
[hours]
Speakers
Key Features
HPSDB
~16
308
Several types of speech, Parkinson’s disease and healthy control. Kiss et al. (2021)
LAPASDA
~7
~400
Read text and sustained sounds; dysphonia and healthy control, RBH scores. Sztahó et al. (2021)
DEPISDA
~17
400
Read text, spontaneous; depression and healthy control, BDI-II scores. Hajduska-Dér et al. (2022)
HuMenDisCo
n.a.
90
Spontaneous; bipolar, schizophrenia and schizoaffective disorder and healthy control. Szabó et al. (2023)
Total approximately
40
1198
Table 3: Hungarian Pathological Speech Data Sets
Name
[hours]
Speakers
Key Features
SPECO
~15
72
5-10 years old children, read speech and sustained segments. Csatári et al. (1999)
SPEECON-2
~10
50
8 to 15-year-old children, home environment, read sentences, toy commands, etc. Iskra et al. (2002)
CHILDES
~13
1
A boy between the ages 1;11 and 2;11, freeplay, home environment. Réger (2004)
Monyek (Eng. HUKILC)
~20
62
4.5–5.5 ages old children, 20–30 minutes of talks. Orosz and Mátyus (2014)
TiniBea
~13
18
8 female, 10 male speakers of 16–18 years, similar protocol with BEA. Gyarmathy and Neuberger (2015)
GABI
~50
~100
3-18 years old speakers, various speech types. Bóna et al. (2019)
Table 4: Hungarian Child-Related Data Sets
Name
[hours]
Speakers
Key Features
MNyHk
13
~110
Interviews from 79 locations in and around Hungary. Vargha (2007)
VIHk
1
18
Interviews from 17 locations in Transylvania (Romania). Fazakas ( 2013 ) ( G á l)
SzMNyHk
3
34
Interviews in Hungarian dialects from Slovakia. Presinszky (2020)
MoMa
18
~100
Interviews in Moldavian (Romanian) Hungarian (Csángó). Eris et al. (2023)
Total approximately
35
262
Table 5: Hungarian Dialectal Speech Data Sets
Figure 1: Types of Hungarian data sets. Left: Length in hours, Right: Number of speakers. MONO: monolingual, MULTI: Hungarian from multilingual collections, P: pathological, CH: child-related, D: dialectal.
Automatic speech recognition (ASR) performs well for high-resource languages with abundant paired audio-transcript data, but its accuracy degrades sharply for most languages due to limited publicly available aligned data. To this end, we introduce WorldSpeech, a 24 kHz multilingual speech corpus comprising 65k hours of aligned audio-transcript data across 76 languages, collected from diverse public sources including parliamentary proceedings, international broadcasts, and public-domain audiobooks. For 37 languages, WorldSpeech provides more than 200 hours of aligned speech, with 28 exceeding 500 hours and 24 surpassing 1k hours. Fine-tuning existing ASR models on WorldSpeech results in an average relative Word-Error-Rate reduction of 63.5% across 11 typologically diverse languages.
Antonis Asonitis, Luca A. Lanzendörfer, Frédéric Berdoz +1
Conversational automatic speech recognition in Hungarian is constrained by the limited amount of publicly available dialogue-style training data. The BEA-Dialogue corpus addresses this need, but its strictly speaker-disjoint train/dev/eval split reduces the usable material to only 85 hours. In this paper, we introduce BEA-Dialogue+, an expanded version of the corpus that relaxes the split criterion for experimenters and dialogue partners while preserving complete separation of the primary speakers. This results in 200 hours of transcribed natural conversations and enables a controlled study of the trade-off between additional training data and speaker overlap across the splits. We evaluate several Whisper- and FastConformer-based models on both corpus versions, including Serialized Output Training (SOT)-based fine-tuning for dialogue transcription. Our results show that the larger corpus is more challenging for models without fine-tuning, whereas SOT-based adaptation yields consistent improvements in WER, CER, cpWER, and cpCER. Overall, BEA-Dialogue+ provides a substantially larger yet still demanding benchmark for Hungarian dialogue ASR, and a practical resource for training and evaluating dialogue transcription systems.
Máté Gedeon, Piroska Zsófia Barta, Péter Mihajlik +1
Department of Telecommunications and Artificial Intelligence, Budapest University of Technology and Economics, Hungary · Speechtex Ltd., Hungary · ELTE Research Centre for Linguistics, Hungary
As large audio language models (LALMs) advance, robust evaluation frameworks have become essential. In this context, Spanish speech understanding under realistic acoustic conditions has received particularly little attention. We introduce ESCUCHA, the first Spanish speech understanding benchmark designed to evaluate LALMs across heterogeneous acoustic conditions and reasoning abilities. ESCUCHA comprises 1,000 human-curated questions paired with audio, totaling 162.9 hours sourced directly ``from the wild'' rather than drawn from existing datasets, with durations ranging from a few seconds to over 80 minutes. The benchmark emphasizes reasoning, spanning 9 perceptual and 10 reasoning categories, and it captures linguistic diversity through multiple Spanish accents and non-normative speech. ESCUCHA further includes multi-audio questions, spoken questions, and audio instructions, and it flags which questions support open-ended evaluation. Benchmarking several state-of-the-art multimodal and speech models reveals substantial performance gaps relative to trained humans.
Fernando López, Ana Ayala, Guillermo Segovia +4
Telefónica Innovación Digital · Universidad Autónoma de Madrid · Universidad Complutense de Madrid