Evolution of Voices in French Audiovisual Media Across Genders and Age in a Diachronic Perspective

Overview

• This paper presents a diachronic (over time) acoustic analysis of the voices of 1023 speakers from French media archives.
• The speakers were categorized based on time period, age group, and gender.
• The fundamental frequency (F0) and first four formants (F1-F4) were measured for each speaker.
• Statistical models were used to examine how the base-F0 (an estimate of vocal register) and vocal tract length changed over time and across genders.

Plain English Explanation

The researchers in this study analyzed the voices of over 1000 people who appeared in French media over several decades. They looked at how certain acoustic features of the voice, like pitch and vocal tract size, changed over time and differed between men and women of different ages.

To do this, they divided the speakers into 32 different categories based on four time periods (the 1950s, 1970s, 1990s, and 2010s), four age groups (20-35, 36-50, 51-65, and over 65), and two genders (male and female). They then measured two key aspects of each speaker's voice:

1. The fundamental frequency (F0), which corresponds to the overall "pitch" of the voice. From the F0 distribution, they calculated the base-F0, which estimates the speaker's vocal register or typical pitch level.

2. The first four formants (F1-F4), which are related to the size and shape of the vocal tract. They used these formant frequencies to estimate the speaker's average vocal tract length.

The researchers then used statistical models to see how the base-F0 and vocal tract length changed over the different time periods and between genders, while accounting for the effects of age.

The results showed that voices tended to get lower in pitch (i.e., have a lower base-F0) over time, independent of gender. They also found that women's voices got lower in pitch as they got older, but men's voices did not change as much with age.

Technical Explanation

This study conducted a diachronic acoustic analysis of the voices of 1023 speakers from French media archives, spanning four time periods (1955/56, 1975/76, 1995/96, 2015/16), four age groups (20-35, 36-50, 51-65, >65), and two genders. The researchers estimated the fundamental frequency (F0) and the first four formants (F1-F4) for each speaker, using procedures designed to ensure high-quality estimations on this heterogeneous data.

From the F0 distribution of each speaker, the base-F0 value was calculated as an estimate of vocal register. The average vocal tract length was also estimated from the formant frequencies. These measures of base-F0 and vocal tract length were then fitted to linear mixed models to evaluate how they changed across time periods and genders, while correcting for age effects.

The results showed a significant effect of time period, with a tendency for voices to become lower in pitch (i.e., have a lower base-F0) over the decades, independent of gender. Additionally, a lowering of pitch with age was observed for female but not male speakers.

Critical Analysis

The researchers acknowledge several limitations in their study. First, the use of archival media data means the audio quality and recording conditions were variable and potentially inconsistent across the different time periods. While the researchers described procedures to mitigate these issues, there may still be some unaccounted-for sources of noise or bias in the acoustic measurements.

Additionally, the study only examined two binary gender categories (male and female), which may not capture the full diversity of gender identity and expression among the speakers. Future research could explore voice passing and non-binary voice gender prediction or the role of language proficiency and F0 entrainment in L2 English speakers.

The researchers also did not investigate other potential factors that could influence voice acoustics, such as measures of acoustic diversity in speech or the development of acoustic models for automatic speech recognition. Incorporating these additional variables could provide a more comprehensive understanding of how voice signal processing and machine learning shape the observed changes in voice acoustics over time.

Conclusion

This study presents a detailed diachronic analysis of voice acoustics in a large dataset of French media speakers. The findings suggest that overall pitch levels have trended lower over the past several decades, with women's voices showing a more pronounced lowering of pitch as they age compared to men. These results contribute to our understanding of how voice characteristics may be evolving, with potential implications for fields like speech technology, linguistics, and social sciences. Future research could build on this work by incorporating additional factors and exploring the voice diversity across a broader spectrum of gender identities.