Formant measurement in children's speech based on spectral filtering

Area of Science:

• Acoustic Phonetics
• Speech Science
• Signal Processing

Background:

• Children's speech presents unique challenges for formant frequency measurement due to high fundamental frequencies and spectral noise.
• Widely spaced harmonics in children's speech can undersample vocal tract characteristics.
• Glottal turbulence contributes to weakened upper harmonic energy, complicating analysis.

Purpose of the Study:

• To develop a formant measurement technique using cepstral analysis for children's speech.
• To avoid modifying the cepstrum or transforming back to the spectral domain.
• To accurately track formant frequencies in pediatric speech signals.

Main Methods:

• A narrow-band spectrum is low-pass filtered using a cutoff "quefrency" to isolate the spectral envelope.
• Speech was simulated using an airway modulation model with physiologically-scaled vocal folds and vocal tract.
• The method was tested on simulated 2-3 year-old child speech, with independently calculated vocal tract resonances serving as ground truth.

Main Results:

• The spectral filtering approach provided a clear spectrographic representation of formant changes over time.
• The method successfully facilitated tracking of formant frequencies in simulated pediatric speech.
• Accuracy was assessed by comparing tracked formants to independently calculated vocal tract resonances.

Conclusions:

• The developed cepstral analysis technique is effective for formant measurement in children's speech.
• This method overcomes limitations associated with high fundamental frequencies and spectral noise in pediatric speech.
• The approach offers a robust tool for analyzing formant dynamics in young children's vocalizations.