Separability of infant-directed from adult-directed speech is affected by number of channels in cochlear-implant simulated speech

Abstract

Many studies have demonstrated benefits of infant-directed speech (IDS) over adult-directed speech (ADS) for language development. For deaf children who use cochlear implants (CIs), a variety of factors might reduce the advantage of IDS in language development. We hypothesized that spectral degradation due to the number of channels in CI processing negatively affects acoustic separability of IDS and ADS, potentially reducing benefits for language development of IDS in early childhood. 493 sentences spoken in two speaking styles (IDS and ADS) were processed using 4, 8, 12, 16, 22, and 32 channels noise-vocoders to simulate the spectral degradation caused by CIs. The sentences were partitioned into frames of 30 ms and represented by Mel-frequency cepstral coefficients. The Mahalanobis distance was used to calculate the acoustic distance across speaking styles (ID versus AD) and processing condition (4, 8, 12, 16, 22, and 32 channels). The results show that spectral degradation imposed by CI processing has significant negative effects on separation of IDS from ADS. These findings suggest that the spectral information in speech received by infants with CIs is substantially degraded, which can weaken the facilitative link between IDS and language development in this at-risk population. [Work supported by NIH grant R01DC008581.]Many studies have demonstrated benefits of infant-directed speech (IDS) over adult-directed speech (ADS) for language development. For deaf children who use cochlear implants (CIs), a variety of factors might reduce the advantage of IDS in language development. We hypothesized that spectral degradation due to the number of channels in CI processing negatively affects acoustic separability of IDS and ADS, potentially reducing benefits for language development of IDS in early childhood. 493 sentences spoken in two speaking styles (IDS and ADS) were processed using 4, 8, 12, 16, 22, and 32 channels noise-vocoders to simulate the spectral degradation caused by CIs. The sentences were partitioned into frames of 30 ms and represented by Mel-frequency cepstral coefficients. The Mahalanobis distance was used to calculate the acoustic distance across speaking styles (ID versus AD) and processing condition (4, 8, 12, 16, 22, and 32 channels). The results show that spectral degradation imposed by CI processing has s...