A Frequency Normalization Technique for Kindergarten Speech Recognition Inspired by the Role of fo in Vowel Perception

Abstract

Accurate automatic speech recognition (ASR) of kindergarten speech is particularly important as this age group may benefit the most from voice-based educational tools. Due to the lack of young child speech data, kindergarten ASR systems often are trained using older child or adult speech. This study proposes a fundamental frequency (fo)-based normalization technique to reduce the spectral mismatch between kindergarten and older child speech. The technique is based on the tonotopic distances between formants and fo developed to model vowel perception. This proposed procedure only relies on the computation of median fo across an utterance. Tonotopic distances for vowel perception were reformulated as a linear relationship between formants and fo to provide an effective approach for frequency normalization. This reformulation was verified by examining the formants and fo of child vowel productions. A 208-word ASR experiment using older child speech for training and kindergarten speech for testing was performed to examine the effectiveness of the proposed technique against piecewise vocal tract length, F 3-based, and subglottal resonance normalization techniques. Results suggest that the proposed technique either has performance advantages or requires the computation of fewer parameters.

Fundamental Frequency Feature Normalization and Data Augmentation for Child Speech Recognition

“…We note that the CMU Kids testing set had a narrower age range (approximately 6-9 years old excluding the two outlier children) compared to the OGI Kids' testing set (approximately 5-11 years old). The f o normalization method has been shown to produce larger improvements when the range of ages used in training and testing data is wider [15]. A similar phenomenon may be occurring for f o perturbation.…”

“…The remainder of the paper is organized as follows. Section 2 reviews the f o normalization technique proposed in [15] and formulates the data augmentation technique. Section 3 describes the databases and experimental setup.…”

“…For instance, vocal tract length normalization (VTLN) warps the frequency spectra using a maximum likelihood approach and can be implemented in several ways with varying degrees of success [10,11,12]. An alternative approach is to use acoustically relevant speech parameters as normalization factors such as subglottal resonances (SGRs) [13], the third spectral peak or formant frequency [14], and f o [15].…”

“…In our previous study, we proposed an f o -based normalization technique for child ASR [15]. In that study, a number of ASR systems were trained using speech from children of various ages and tested using kindergarten-aged children.…”

“…In this study, we propose a new data augmentation method for training child ASR systems based on the f o normalization method we proposed in [15]. This method is capable of generating speech-like features that adhere to the physical properties of speech defined by the relationship between vowel formants and f o .…”