Code for acoustic distance (AD) calculation between languages, AD-based evolutionary tree generation, dissimilarity metric calculation between trees, and AD-based chronology estimation; as well as data generated from these calculations on 42 Indo-European languages.

The dataset used as the example here, synthesized audio of 300 core words of 42 Indo-European languages, can be retrieved from Github or Zenodo. Please place the dataset folder acoustic-dist-ie-audio in the parent directory of this repository.

For convenience, it is highly recommended to download the single-file archived dataset from Github or Zenodo instead of downloading the original dataset. Please put the archived dataset file audios.joblib in the folder of this repository.

You can also create new datasets to conduct AD-based investigations of phylogeny and chronology for other languages.

Some Python libraries need to be installed for acoustic feature (AF) extraction, DTW calculation, and clustering.

Run acoustic_distance.py to extract AF and compute AD. AF, AF normalization method, DTW method, and AD normalization method can be selected via function parameters. Audio metadata is given in audio_metadata.py. Resulting distance matrices will be saved as .csv files in acoustic_distances. When run for the first time, the archived audio dataset file audios.joblib mentioned above will be generated (if it does not exist) in order to speed up the calculation process.

Since the calculation process is quite time-consuming, please comment out the parameters you do not want to extract in acoustic_distance.py (Lines 36–46) to save time. Besides, AD matrices under all parameter combinations are stored in this repository in acoustic_distances, so you can quickly skip the AD calculation process and run the following process directly.

Run clustering.py to generate AD-based evolutionary trees from AD matrices in acoustic_distances. Clustering method can be selected via function parameters. Results in Newick format will be saved in trees/newicks.tsv.

R packages Quartet and TreeDist need to be installed for dissimilarity metric calculation.

Run dissimilarity_metric.r in R to compute the Steel–Penny metric and the Robinson–Foulds metric between AD-based trees in trees/newicks.tsv and the reference tree trees/reference_tree.nwk. Results will be saved as trees/dissimilarity_metrics.csv.

The reference tree here is a hierarchy of 42 Indo-European languages sourced from Glottolog accompanying the dataset.

Run fitting.py to fit AD and date, and estimate chronology from the AD. Branch and date data of calibration and prediction points are given in fitting_data.py. Results will be printed.

• acoustic_distances/: AD matrices under all parameter combinations
• trees/newicks.tsv: Strings of all AD-based clustering trees in Newick format
• trees/reference_tree.nwk: The reference tree in Newick format
• trees/dissimilarity_metrics.csv: Dissimilarity metrics of all trees