Automated Relative Fundamental Frequency Algorithms for Use With Neck-Surface Accelerometer Signals.

Area of Science:

• Speech Science
• Acoustic Analysis
• Biomedical Engineering

Background:

• Relative fundamental frequency (RFF) is a potential acoustic measure for vocal effort.
• Current clinical RFF measurement methods are time-consuming due to manual annotation.
• Existing semi-automated algorithms for RFF calculation rely on microphone signals.

Purpose of the Study:

• To develop fully automated algorithms for calculating RFF from neck-surface accelerometer signals.
• To enable ecological momentary assessment and ambulatory monitoring of voice.
• To overcome limitations of manual RFF measurement in clinical settings.

Main Methods:

• Developed automated algorithms using a training set of 2646 /vowel-fricative-vowel/ utterances from 317 speakers.
• Algorithms included steps for rejecting poor signal-to-noise ratios, identifying fricative locations, and determining voicing boundaries.
• Validated automated RFF values against the clinical gold-standard (manual RFF from microphone signals) using a test set of 639 utterances from 77 speakers.

Main Results:

• Automated accelerometer-based RFF values showed an average mean bias error (MBE) of 0.027 ST across all cycles.
• Specific MBEs were 0.152 ST and -0.252 ST for offset and onset cycles near the fricative, respectively.
• These errors are smaller than expected changes in RFF following voice therapy.

Conclusions:

• The developed fully automated algorithms provide accurate RFF measurements from neck-surface accelerometer signals.
• These algorithms are suitable for ecological momentary assessment and ambulatory voice monitoring.
• This technology offers a more efficient and accessible approach to voice analysis.