Enhancing Children’s Short Utterance-Based ASV Using Inverse Gamma-tone Filtered Cepstral coefficients

The task of developing an automatic speaker verification (ASV) system for children’s speech is extremely challenging due to the dearth of domain-specific data. The challenges are further exacerbated in the case of short utterances of speech, a relatively unexplored domain in the case of children’s ASV. Voice-based biometric systems require an adequate amount of speech data for enrollment and verification; otherwise, the performance considerably degrades. It is for this reason that the trade-off between convenience and security is gruelling to maintain in practical scenarios. In this paper, we have focused on data paucity and preservation of the higher-frequency contents in order to enhance the performance of a short utterance-based children’s speaker verification system. To deal with data scarcity, an out-of-domain data augmentation approach has been proposed. Since the out-of-domain data used are from adult speakers which are acoustically very different from children’s speech, we have made use of techniques like prosody modification, formant modification, and voice conversion in order to render it acoustically similar to children’s speech prior to augmentation. This helps in not only increasing the amount of training data but also in effectively capturing the missing target attributes which helps in boosting the verification. Further to that, we have resorted to concatenation of the classical Mel-frequency cepstral coefficients (MFCC) features with the Gamma-tone frequency cepstral coefficient (GTF-CC) or with the Inverse Gamma-tone frequency cepstral coefficient (IGTF-CC) features. The feature concatenation of MFCC and IGTF-CC is employed with the sole intention of effectively modeling the human auditory system along with the preservation of higher-frequency contents in the children’s speech data. This feature concatenation approach, when combined with data augmentation, helps in further improvement in the verification performance. The experimental results testify our claims, wherein we have achieved an overall relative reduction of 38.5% for equal error rate.