Plotting Pitch and Intensity Contours with Annotations in Praat

Graphs export via export button or ggsave(). Working directory set with setwd(). ggsave() parameters include filename, plot, width, height, dpi, limitsize, and background. Formats include PNG, JPG, PDF, and TIFF. PDF and PNG are preferred for publication due to quality and file size balance. The tidyverse provides tools for data transformation. Filter() selects observations using %in% operator. Group_by() creates grouped data structures for category-based operations. Summarize() calculates statistics per group. These functions enable transforming raw data into analysis-ready formats.

This section covers the practical implementation of ggplot2 and R project management. The ggplot2 syntax follows: ggplot(data) + aes(aesthetics) + geometry + theme. Alternatively, ggplot(data, aes(aesthetics)) + geometry + theme can be used. The 'gg' in ggplot2 is always lowercase. The export() function from the 'here' package is used to save outputs. Project environment management is explained: when you close and reopen your R environment, it becomes empty, and scripts are stored in your R folder. Reopening the environment allows you to continue your work.

This comprehensive section establishes the fundamental skills required for analyzing acoustic telemetry data using R. It covers: tidyverse data manipulation with pipes, select(), filter(), and arrange(); date-time handling with lubridate; spatial data conversion using sp::coordinates() and sp::transform(); and marine mapping with mapview. The workflow progresses from basic data inspection to creating bathymetric maps, establishing the complete pipeline for handling and visualizing marine telemetry datasets.

R provides powerful data visualization through base R graphics and ggplot2. Base R uses plot() for scatter plots, hist() for histograms, and boxplots for quartile displays. Save graphics using PDF() for vector files (preferred for academic papers) or BMP/PNG for lossless formats. ggplot2 implements the grammar of graphics, building plots through layered components: data, aesthetics (aes()), and geoms (geom_point, geom_boxplot, geom_jitter). Themes like theme_bw() control appearance. Map variables to aesthetics (color, fill, size, shape, alpha) to encode information. For summary statistics, use geom_boxplot() or pre-calculate in dataframes. For model-based predictions, use emmeans() to create prediction dataframes. Combine raw data with summary statistics by using separate dataframes for each geom. Order geoms carefully: background elements first, foreground elements last.

In R, plots created with ggplot2 can be exported as images using the ggsave() function by specifying the filename with an appropriate extension (e.g., .jpg, .tiff, .pdf, .svg), and plot dimensions can be controlled using width, height, and units parameters; bitmapped formats like JPEG and TIFF store pixel-based images suitable for web use, while vector formats like PDF and SVG preserve quality at any scale and allow individual element modification in graphic design software.

English intonation operates as a supra-segmental feature that superimposes pitch contours on top of word-based stress patterns, where pitch accents (H* and L*) align with stressed syllables to create prominence, and boundary tones (%H and %L) appear at phrase ends; the TOBI transcription system combines these pitch accents with break indices (1 and 4) to mark word boundaries and phrase boundaries respectively, allowing for systematic analysis of how intonation conveys pragmatic and discourse information in English speech.

This video presents an alternative model for representing sentence intonation that replaces the traditional ToBI notation system (which uses high/low tone targets) with a melodic contour approach based on the glissando threshold. The glissando threshold formula (st2-st1)/(t2-t1) separates perceived melodic changes from static tones, creating four contour classes: rising/falling above the threshold (melodic) and rising/falling below the threshold (neutralized). This approach addresses fundamental problems in ToBI, including the confusion between phonetic and phonological transcription, the arbitrary treatment of insignificant pitch changes, and the lack of experimental backing for tonal targets in non-tonal languages. The model proposes that prosodic structure should be defined by dependency relations between melodic contours rather than as an interface with syntax, with neutralization processes playing a central role in the grammar of intonation.

This lecture introduces seven methods for visualizing and representing prosodic information in speech: (1) Pitch contours with blue dots showing individual pitch points; (2) Smoothing techniques to create continuous curves from discrete points; (3) Pitch turning points extraction to identify key changes; (4) Tadpole notation using height and circle size to mark pitch and loudness syllable-by-syllable; (5) ToBI notation categorizing pitch as high (H) or low (L) and focusing on stressed syllables; (6) Ribbon diagrams encoding both pitch and intensity; (7) Word-level abstractions showing average pitch per word. The lecture emphasizes two strategies: tidying up (smoothing and interpolating) and extracting what matters (identifying perceptually salient features). Key cautions include recognizing that auditory and visual perception may differ, intonation alone can be misleading, and model assumptions may not apply universally across languages.

There is a significant disagreement among linguists about how to analyze English intonation. The British school traditionally views intonation in terms of pitch movement patterns, identifying five main intonational patterns where the most important word in a sentence carries a specific pitch contour (fall for declaratives, rise for questions). The American school uses the Tones and Break Indices (ToBI) system, which focuses on target tones (high or low relative to surrounding speech) rather than the melodic contours between them.

The PROPER toolbox provides open-source tools for analyzing speech prosody using periodic energy. Key metrics include: (1) Mass—syllable weight measured as area under the periodic energy curve, combining duration and power; (2) Delta f0—pitch differences between consecutive syllables measured at centers of mass; (3) Synchrony—characterizing f0 contour shape by relating center of mass to center of gravity. Mass outperforms separate duration/intensity measures for detecting prominence and focus, as demonstrated in studies on German and Italian intonation.

This section covers automating formant extraction using Praat scripts that analyze sounds with the Berg algorithm and extract formant values at multiple time points within annotated vowel intervals. Convert custom codes to IPA symbols for professional notation. Use R's ggplot2 to plot vowel trajectories with F2 on horizontal and F1 on vertical axes (reversed for standard charts). Apply smoothing techniques and exclude first/last time points to avoid consonant influence. This creates dynamic vowel charts showing formant movement throughout vowel duration.

Formants are the centers of resonant frequencies in a vowel sound, representing the loudest frequencies that determine how we perceive that vowel. The first formant (F1) corresponds to vowel height - higher F1 values indicate more open vowels. The second formant (F2) corresponds to vowel frontness or backness - higher F2 values indicate more fronted vowels. These two formants can be visualized on a graph where F1 is plotted on the Y-axis and F2 on the X-axis, creating the familiar IPA vowel trapezoid diagram.

To measure formant frequencies, select a stable middle portion of a vowel, place cursors on the darkest regions of formant bars, and record F1 and F2 values. Plot these on a chart with F1 on the vertical axis (representing height) and F2 on the horizontal axis (representing front-back position). Lower F1 indicates higher vowels; higher F2 indicates fronter vowels. This plotting technique allows visualization of vowel quality differences across speakers and helps identify articulatory patterns from acoustic data.

This section covers formants and how spectrograms visualize sound. A spectrogram displays frequency over time: horizontal axis is time, vertical axis is frequency (lower at bottom, higher at top), and brightness represents amplitude. Formants are peaks in amplitude at specific frequencies, appearing as dark patches in spectrograms. They are numbered from lowest to highest: F1 (first formant, lowest frequency), F2 (second formant), F3 (third formant), and so on. The lowest two formants (F1 and F2) determine vowel quality (whether a sound is 'ah', 'ee', or 'oh'). Higher formants determine voice timbre, which distinguishes one person's voice from another's. This explains how two people can have different voices but produce the same vowel quality.

Vowel formants are resonant frequencies of the vocal tract that determine vowel quality; the first formant (F1) corresponds inversely to vowel height (higher F1 means lower tongue position), while the second formant (F2) corresponds to vowel advancement (higher F2 means more front tongue position). These formants can be visualized in spectra as dark energy bands or in spectrograms over time, and plotting F1 against F2 produces a vowel chart that systematically represents all vowel qualities based on their acoustic measurements.