Title: BabyEars Affective Vocalizations


Authors: Malcolm Slaney, Gerald McRoberts, Jocelyn Scheirer


Language: English


Expected use of corpus: 
This dataset will be useful for researchers studying how adults convey 
emotional messages in speech.  Not big enough for modern training methods,
it is useful as a historical baseline for emotion recognition experiments.


Collection procedure: 
Adults were asked to speak naturally to their infant babies while they explored
a room. They were asked to prohibit with their voices the infant from touching 
"dangerous" items.  After the recording/play session the emotional and neutral 
utterances were segmented by a trained experimenter.  Note: these natural
recordings are in contrast to other datasets that use trained actors to interpret
the directions.

See more details in this paper:
Malcolm Slaney, Gerald McRoberts.  BabyEars: A recognition system for affective 
vocalizations. Speech Communication 39 (2003) 367–384.


Directory Structure:
The sound files in .WAV format are stored in the directory with names that look
like: 
	subSSclip/SS.0?.CC.NN.wav
where SS is a subject number between 1 and 12.  CC is the emotional class
(ad: Adult directed or neutral, ap: Approval, at: Attention, pr: Prohibition),
NN is the utterance number.

All data is stored as 16 bit FLAC files at 22 kHz sampling rate