Using automatic speech processing to analyze fundamental frequency of child-directed speech stored in a very large audio corpus

Child-Directed-Speech (CDS) is associated with raised fundamental frequency (f₀). In a previous paper we claimed that f0 could be extracted from 500 hours of audio recordings using soft computing techniques and that mothers, but not fathers, increase f₀ in CDS. Using an audio corpus more than ten times larger, this paper reports that fathers do raise f₀ but not as much as mothers. The principle finding is a proof of concept: 1) very large speech corpora, unavailable until recently, can be processed using soft computing techniques; 2) the use of very large corpora may force revisions of conclusions based on smaller datasets.