Automated Relative Fundamental Frequency Algorithms for Use With Neck-Surface Accelerometer Signals.
Matti D Groll, Jennifer M Vojtech, Surbhi Hablani and 4 others
PMID 32653267WHAT IT FOUND
Automated algorithms estimated relative fundamental frequency from neck accelerometer signals with errors similar to microphone algorithms.
Offset values were more accurate than onset values, and results should be averaged across multiple utterances.
Key findings
01In an independent test set of 639 VCV utterances from 77 speakers, the automated accelerometer algorithms had errors comparable to previous microphone algorithms.
02The algorithms were more accurate for offset cycles than for onset cycles, and onset 1 errors were larger than those of both microphone algorithms.
03The algorithms should be used to calculate average RFF estimates across multiple utterances from the same speaker, not to rely on individual utterances.
STILL TO COME
How it was doneWhat they foundWhat it means for SLPs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The study evaluated algorithm performance on recorded speech samples, not on patients receiving voice therapy, so it does not show that the algorithms improve treatment decisions. The algorithms were tested only on isolated /afa/, /ifi/, and /ufu/ utterances, not on running speech, so they cannot yet monitor everyday voice use. All data were collected with one Knowles accelerometer, so performance may differ with other devices or setups. The test set rejected 312 of 639 utterances for signal quality, and many participants had few usable utterances, with 5.0 offset and 4.7 onset utterances on average. Onset estimates were less accurate than offset estimates, especially onset cycle 1, so onset values may not be useful for monitoring vocal strain yet. The study did not compare device performance separately for speakers with and without voice disorders because signal quality was confounded.
The easy way to misread this
Do not read the comparable average errors as evidence that neck-surface accelerometer RFF can replace microphone-based clinical RFF. The algorithms rejected many utterances, onset values were less accurate, and the paper states that microphone-based manual RFF remains the gold standard.