Beszédkorpuszok

Beszédkorpuszok, Hungarian for speech corpora, are structured collections of recorded spoken language that are used for linguistic research, language technology development, and educational purposes. Each corpus typically contains audio recordings accompanied by transcriptions, phonetic annotations, and metadata such as speaker identity, gender, age, and speaking context. The most common types of speech corpora include read speech, in which speakers read prepared texts, conversational speech, comprising spontaneous dialogues between participants, and broadcast corpora that document radio, television, and online media content.

The earliest Hungarian speech corpora emerged in the 1990s, with the Hungarian Speech Corpus (HSC) providing thousands of hours of read and conversational data. More recent projects, such as the Transcribed Hungarian Corpus of Spoken Language (THCoSL), have expanded the linguistic coverage and incorporated multimodal annotations like gesture and prosody. Speech corpora have become essential resources for developing acoustic models for automatic speech recognition (ASR) and for phonetic studies of Hungarian prosody and phonotactics.

Collection methods vary: some corpora are recorded in controlled studio environments using high‑fidelity microphones, while others are harvested from naturalistic settings such as call centers or news broadcasts. Digital recordings are often stored in WAV or FLAC format, with accompanying annotation files in TextGrid, XML, or ELAN format. Researchers and developers can access many Hungarian corpora through open‑access repositories like the Hungarian Linguistic Data Center (HLDC), ensuring that the data can be used for both academic and commercial applications.

The continued development and refinement of beszédkorpuszok support advancements in speech technology, linguistic theory, and language preservation, making them a cornerstone of contemporary Hungarian language studies.