Äänikorpora

Äänikorpora, literally “audio corpora” in Finnish, are collections of recorded speech and related annotations used in linguistics and speech technology. They combine sound recordings with time-aligned transcripts and often multiple annotation layers to enable phonetic, prosodic, and pragmatic analyses as well as the development of automatic speech recognition (ASR) and speech synthesis systems.

A typical äänikorpora includes audio data (commonly WAV or FLAC), orthographic transcripts, and alignment information such as word or phoneme timings. Additional metadata may cover speaker attributes (age, sex, dialect), recording conditions, and equipment. Annotation layers can include phonetic transcriptions, part-of-speech tags, prosody, and discourse markers, stored in formats like TextGrid, ELAN, or structured JSON/XML.

Creation and rights: materials are gathered with informed consent and managed under data protection and privacy rules. Accessibility varies by corpus: some are open access, others require agreements or are restricted to research purposes or specific institutions.

Uses: äänikorpora support research on Finnish phonetics and prosody, dialectology, language documentation, and the development and evaluation of ASR and TTS systems. They also serve educational and archival purposes.

Format and standards: common media are lossless audio alongside human- or machine-readable annotations. Alignments are time-stamped and synchronized with transcripts, enabling precise analysis of speech timing and intonation.

Size and availability: äänikorpora range from tens of hours to thousands of hours of speech, depending on scope and funding. Access policies and licensing determine reuse and distribution.