![enter image description here][1] # The overlooked effect of amplitude on within-speaker vowel variation ## Supplementary material and data Wilson Black, Joshua, Jen Hay, Lynn Clark, and James Brand. _Linguistics Vanguard_ (2023). DOI: https://doi.org/10.1515/lingvan-2022-0086 [<img src="https://img.shields.io/badge/GitHub-repo-blue?labelColor=lightgrey&logo=github">](https://github.com/nzilbb/amp_f1_public) --- ### Abstract We analyse variation in vowel production within monologues produced by speakers in a quiet, well-controlled environment. Using Principal Component Analysis and Generalized Additive Mixed Models, applied to a large corpus of naturalistic recordings of New Zealand English speakers, we show that the first formant of monophthongs varies significantly with variation in relative amplitude. We also find that amplitude variation is used, potentially agentively, to mark the beginning and ending of topical sections within single-speaker monologues. These results have significant methodological consequences for the study of vocalic variation in the context of research on speaker style, and language variation and change. While laboratory research has shown a connection between variation in F1 and amplitude in loud environments or with distant interlocutors, this has not been seen in quiet environments with unscripted speech of the sort often used in sociolinguistics. We argue that taking account of this variation is an important challenge for both within speaker investigation of stylistic covariation and across speaker investigation. In the latter case we recommend, as a minimal step, the inclusion of a measure of relative amplitude within regression models. ### How to use this repository This repository contains R code, commentary and anonymised data. The code and commentary comes in the form of four R markdown files which cover: **1.** data processing, **2.** dividing monologues into intervals, **3.** application of PCA to interval means, and **4.** application of GAMM and GLMM models. The simplest way to use this material is to view the supplementary files using your web browser. The benefit of this is that you can easily see the commentary and the code and do not have to run it yourself (which takes a long time). The disadvantage is, correspondingly, that you cannot step through each step in your own R session or see what happens if you modify the code. The web versions of the supplementary materials are available at: 1. [Preprocessing](https://nzilbb.github.io/amp_f1_public/supplementary_material/SM1_preprocessing.html) 2. [Interval representation](https://nzilbb.github.io/amp_f1_public/supplementary_material/SM2_interval_representation.html) 3. [PCA](https://nzilbb.github.io/amp_f1_public/supplementary_material/SM3_corpus_pca.html) 4. [Models](https://nzilbb.github.io/amp_f1_public/supplementary_material/SM4_models.html) To interact with the code yourself, there are two necessary steps: 1. **Acquire code from GitHub repository:** If you are comfortable with git, clone the [GitHub repository](https://github.com/JoshuaWilsonBlack/amp_f1_public) to your own machine. If you are not, then download the code as a ZIP file [here](https://github.com/JoshuaWilsonBlack/amp_f1_public/archive/refs/heads/main.zip). 2. **Acquire data (and/or models) from this OSF repository.** The anonymised data is too large for a standard GitHub repository and so is stored here at OSF as a '.rds' file, which is openable by R using the base R `readRDS` function or `readr::read_rds`. The models reported in the paper and supplementary material are also available in OSF storage. Place models and data in the relevant folders within the repository (`labbcat_data` for the data file and `models` for the models). [1]: https://github.com/nzilbb/Covariation_monophthongs_NZE/raw/master/Covariation_shiny/www/NZILBB2.png