Age-related changes in fundamental frequency and formants: a longitudinal study of four speakers

Jonathan Harrington, Sallyanne Palethorpe, Catherine I. WatsonInterspeech 2007

TL;DR

This longitudinal study analyzes acoustic changes in four speakers over 29-50 years, revealing consistent age-related decreases in fundamental frequency (f0) and first formant (F1). The findings suggest these shifts, while mimicking broader dialectal changes, are likely due to physiological vocal tract changes rather than conscious phonetic shifts.

What Problem Does This Paper Address?

Understanding how our voices change as we age is crucial for fields ranging from speech synthesis to forensic phonetics. While previous research has often relied on cross-sectional studies comparing different age groups, this paper tackles the challenge of tracking individual vocal changes over extended periods. The authors aim to determine if observed age-related acoustic shifts are genuine physiological changes within a speaker or if they reflect broader, potentially conscious, phonetic trends within a dialect. They specifically investigate changes in fundamental frequency (f0) and formant frequencies (F1, F2, F3), which are key acoustic correlates of vowel quality and vocal tract characteristics.

Methods

The study employs a longitudinal acoustic analysis of recorded speech from four individuals. The core methodology involves:

  • Data Collection: Recordings spanning 29 to 50 years were collected from four speakers: Queen Elizabeth II, actress Margaret Lockwood, an Australian English male speaker, and a New Zealand English male speaker. These recordings were primarily from public broadcasts (Christmas broadcasts, radio programs).
  • Acoustic Analysis: Speech data was digitized, and the first four formant frequencies (F1-F4) and fundamental frequency (f0) were extracted using an LPC-based autocorrelation method with a Blackman window. Analysis was performed on all voiced frames and specifically on schwa sounds at their temporal midpoint.
  • Statistical Analysis: Comparisons were made between early and late recordings using t-tests to determine the statistical significance of observed changes in f0 and formants. Percentage changes were also calculated to quantify the magnitude of these shifts.
  • Perceptual Space Modeling: The authors utilized Bark-scaled formant differences (F1-f0 in Bark and F3-F2 in Bark) to model changes in perceived phonetic height and backness, aiming to link acoustic shifts to potential perceptual consequences.

Datasets

The study utilized existing broadcast recordings from four individuals:

  • Queen Elizabeth II: Annual Christmas broadcasts spanning the 1950s to the 2000s.
  • Margaret Lockwood: Recordings from the BBC radio program "Desert Island Discs" from 1951 and 1980.
  • Australian English Male Speaker: Broadcasts from 1948 and 1987.
  • New Zealand English Male Speaker: Recordings from 1955 and 1992.

These datasets represent longitudinal recordings of native speakers of different English dialects, providing a rich basis for tracking individual vocal changes over decades.

Key Insights and Ideas

  • Consistent f0 and F1 Decrease: A striking and consistent finding across all four speakers is the decrease in both fundamental frequency (f0) and the first formant (F1) from earlier to later recordings. This suggests a fundamental physiological change in the vocal apparatus with age.
  • F2 Tends to Fall, F3 is Inconsistent: While F2 also showed a tendency to decrease, the changes in F3 were less consistent, with some speakers showing a slight increase. This differential change across formants is key to understanding the overall vocal tract configuration.
  • Mimicking Dialectal Shifts: The observed age-related shifts in vowel formants (e.g., F1 and F2 lowering) were found to be in a similar direction to broader phonetic changes observed in dialects like Received Pronunciation (RP) and New Zealand English. This raises the question of whether individual aging mimics societal sound change.
  • Physiological vs. Phonetic Explanation: The authors strongly argue against a conscious phonetic shift explanation. The consistency of f0 and F1 changes across different speakers and dialects, and the fact that these changes are not always perfectly aligned with known dialectal shifts, points towards passive physiological changes in the vocal tract (e.g., lengthening) as the primary driver.
  • Perceptual Constancy Hypothesis: The paper proposes that these age-related acoustic shifts might be a mechanism for maintaining perceptual vowel quality. As the vocal tract changes, f0 and F1 decrease, potentially keeping the perceived vowel height constant for listeners.
  • Diagonal Shift in Speaker Space: Modeling changes in Bark-scaled formants suggests that the "speaker space" (representing phonetic height and backness) shifts diagonally with age. This implies that while the acoustic parameters change, the relationship between them might be maintained, leading to a phonetically neutral outcome for listeners.

Key Results

  • f0 Decrease: Queen Elizabeth II's f0 decreased by 59 Hz (22.1%) from the 1950s to the 2000s. Similar significant decreases were observed in the other three speakers.
  • F1 Decrease: Queen Elizabeth II's F1 decreased by 116 Hz (21.9%) over the same period. Again, this was a consistent finding across all speakers.
  • F2 Decrease: A tendency for F2 to fall was observed, with Queen Elizabeth II's F2 decreasing by 80 Hz (4.5%).
  • F3 Inconsistency: F3 showed less consistent changes, with a small rise of 29 Hz (1.0%) for Queen Elizabeth II, but the overall trend was not statistically significant across decades for her.
  • Statistical Significance: T-tests confirmed statistically significant differences in f0, F1, and F2 between early and late recordings for most speakers, while F3 changes were often not significant.
  • Bark Scale Analysis: The analysis in the Bark scale revealed that while F1 and f0 converged (decreased), F3 and F2 diverged, suggesting a shift in the speaker's phonetic space.

Limitations and Future Work

  • Limited Number of Speakers: The study is based on only four speakers, which limits the generalizability of the findings. While these speakers represent different genders and dialects, a larger sample size would be beneficial.
  • Broadcast Data Limitations: The use of broadcast data, while providing long-term recordings, may introduce variability due to speaking style, microphone differences, and recording quality across different eras.
  • Passive vs. Active Modification: The paper acknowledges that it remains an open question whether these shifts are purely passive physiological changes or if speech production is actively modified by speakers to compensate for age-related vocal decline. Future research could explore this active compensation aspect.
  • Perceptual Validation: While the study proposes a perceptual explanation for the acoustic shifts, direct perceptual experiments would be needed to confirm whether listeners indeed perceive phonetic constancy despite these age-related vocal changes.
  • Broader Dialectal and Gender Comparisons: Further research could expand on comparing these individual longitudinal changes with broader cross-sectional studies across more dialects and genders to better understand the interplay between individual aging and societal sound change.

Summary updated

BibTeX
@inproceedings{harrington2007age,
  author = {Jonathan Harrington and Sallyanne Palethorpe and Catherine I. Watson},
  title = {{Age-related changes in fundamental frequency and formants: a longitudinal study of four speakers}},
  booktitle = {Interspeech},
  year = {2007},
}
Figure