Cluster f0 contours into candidate tone categories

Usage

cluster_f0(
  feat,
  method = c("kmeans", "hclust", "gmm"),
  k,
  nstart = 25L,
  hclust_method = "ward.D2"
)

Arguments

feat: A list from cluster_features() (needs features+contours).
method: "kmeans", "hclust" (Ward), or "gmm" (model-based, needs mclust).
k: Number of clusters.
nstart: k-means restarts. Default 25.
hclust_method: Linkage for "hclust". Default "ward.D2".

Value

A list with assignment (named integer per token), method, k, sizes, cluster_means (k x n_points mean contour per cluster), tokens, uncertainty (per-token 1 - max posterior, GMM only; else NULL), and tree (the fitted hclust object for method "hclust", else NULL).

References

Kaland, C. (2023). Contour clustering: A field-data-driven approach for documenting and analysing prototypical f0 contours. Journal of the International Phonetic Association, 53(1), 159–188. doi:10.1017/S0025100321000049

Hartigan, J. A., & Wong, M. A. (1979). Algorithm AS 136: A k-means clustering algorithm. Applied Statistics, 28(1), 100–108.

Ward, J. H. (1963). Hierarchical grouping to optimize an objective function. Journal of the American Statistical Association, 58(301), 236–244.

Scrucca, L., Fop, M., Murphy, T. B., & Raftery, A. E. (2016). mclust 5: Clustering, classification and density estimation using Gaussian finite mixture models. The R Journal, 8(1), 289–317.