Resource: CAREGIVER Corpus

Reference CAREGIVER Corpus
Date of Submission Sept. 3, 2020, 4:14 p.m.
Status accepted
ISLRN 072-357-063-759-1
Resource Type Primary Text
Media Type Audio
Source
Language Dutch, Flemish, English, Finnish
Size 66000 utterances
Description

A multi-lingual speech corpus used for modeling language acquisition called CAREGIVER has been designed and recorded within the framework of the EU funded Acquisition of Communication and Recognition Skills (ACORNS) project. The motivation behind the corpus and its design relies on current knowledge regarding infant language acquisition. Instead of recording infants and children, the voices of their primary and secondary caregivers were captured in both infant-directed and adult-directed speech modes over four languages in a read speech manner. The challenges and methods applied to obtain similar prompts in terms of complexity and semantics across different languages, as well as the normalized recording procedures employed at different locations, are covered. An orthographic transcription is available for every utterance. Also, time-aligned word and phone annotations for some of the sub-corpora exist.

However, in the actual corpus there are a couple of deviations from this setup. The corpus contains nearly 66,000 utterance-based audio files spoken over a two-year period by 16 male and 14 female native speakers of Dutch, English, and Finnish. Swedish is not provided. For Dutch only year 2 recordings are available.

Overview:

1) UK English:
Year 1:
- 4 speakers (2 males, 2 females)
- 1000 recordings per speaker
- orthographic transcriptions in .xml and speech recordings in .wav
Year 2 :
- 10 speakers including 4 speakers (same as for year 1) with 2397 recordings per speaker and 6 speakers (3 males, 3 females) used as test speakers with 600 recordings per speaker
- orthographic transcriptions in .xml and speech recordings in .wav
- annotation: time stamps at word and phone levels by Forced Alignment and a list of errors in time stamps at word level

2) Finnish:
Year 1 :
- 4 speakers (2 males, 2 females)
- 2000 recordings per speaker
- orthographic transcriptions in .xml and speech recordings in .wav
Year 2:
- 10 speakers including 4 speakers (same as for year 1) with 2397 recordings per speaker and 6 speakers (3 males, 3 females) used as test speakers with 600 recordings per speaker
- orthographic transcriptions in .xml and speech recordings in .wav

3) Dutch:
Year 2:
- 10 speakers including 4 speakers recorded twice (2 males and 2 females) and 6 speakers (4 males and 2 females) used as test speakers with one recording session.
- orthographic transcriptions in .cor and speech recordings in .wav
- annotation: time stamps at sentence level only

To be mentioned as reference to the corpus:
Altosaar, T., Bosch, L. ten, Aimetti, G., Koniaris, Chr., Demuynck, K., Heuvel, H. van den (2010): A Speech Corpus for Modeling Language Acquisition: CAREGIVER. Proceedings LREC2010, Malta, pp. 1062-1068. http://www.lrec-conf.org/proceedings/lrec2010/pdf/597_Paper.pdf.

Version 1.0
Distributor ELRA