Acoustic variability and automatic recognition of children’s speech

Adaptive feature truncation to address acoustic mismatch in automatic recognition of children's speech

“…Also, for ASR of adults' speech, the improvements obtained are less than that for ASR of children's speech under mismatched conditions. This is because the variances for the observation densities of the phone models are greater for the poor models trained on child speech than for the models trained on adult speech [2,8]. This means that the Gaussian densities are more scattered and thus less separable in the acoustic feature space for models trained on child speech.…”

“…VTLN and CMLLR are the two effective techniques in the literature that are used to reduce acoustic mismatch between adult speech and child speech [8,11]. Our proposed algorithm for adaptive MFCC feature truncation also addresses acoustic mismatch and has given significant improvements in performance.…”

“…as maximum a posteriori and maximum likelihood linear regression (MLLR) adaptations [8], constrained MLLR (CMLLR) adaptation [8,10], speaker adaptive training (SAT) [10], constrained MLLR-based speaker normalization [11], and their combinations [8]. Significant improvements have been reported in ASR performance on children's speech under mismatched conditions using each of these speaker adaptation methods.…”

Towards age-independent acoustic modeling

“…In (McGowan and Nittrouer, 1988;Nittrouer and Whalen, 1989;Lee et al, 1999;Narayanan and Potamianos, 2002;Gerosa et al, 2007), it was shown that acoustic and linguistic characteristics of children's speech are widely different from those of adults. Furthermore, these studies also show that characteristics of children's speech vary rapidly as a function of age due to the anatomical and physiological changes occurring during a child's growth and because children become more skilled in coarticulation with age.…”

Conversational Characters that Support Interactive Play and Learning for Children

“…These research findings motivated us to collect a corpus of children conversational data. In fact, the few existing corpora of children speech turned out to be not usable in our system for none of them was in Danish and moreover consisted of either prompted speech or monologues of children recounting stories (D'Arcy et al, 2004;Eskenazi, 1996;Gerosa & Giuliani, 2004;Hagen et al, 1996). We transcribed and analyzed several hours of collected video and audio-taped conversation of young subjects involved in a series of interactive sessions in both Wizard of Oz studies and in an after-school class where they played with a real human actor impersonating Hans Christian Andersen (Figure 3, left).…”