tjmahr
commented
6 months ago I think this should be a separate function. The hard part is coming up with a name for it. Right now, I have pivot_nested_intervals(). I don't want to say pivot_wider or some variation with the word wider because I don't want to interfere with tidyr::pivot_wider() functions.
Here is sketch of it so far:
pivot_nested_intervals <- function(data, id_cols, nesting) {
{
# assertions/validation
# - that only `tier_names` are in `nesting`
# - point tiers can only occur in last layer in nesting
}
# split in to a list for each level of nesting
l <- data |> split(~tier_name) |> _[nesting]
# prepare each list element for table joins, renaming columns, computing midpoints
{}
# for each pair in nesting i, i+1 {
# left_join(l[[i]], l[[i+1]], join_by(id_cols, ..., between(y$xmid, x$xmin, x$xmax)), relationship = "one-to-many")
}
tail(l)
}
So that for your example, you start with the sentence rows and join on all the word rows where the midpoint of the word interval falls between() the boundaries of the sentence interval. You now have one row per word. Repeat that process, join to each word all the ph intervals with a midpoint between() the boundaries of the word interval.
There is no validation that a ph interval is full contained within its parent word interval. I don't want to make that assertion because I have textgrids where the word/phone intervals are adjusted but not forced to exactly line up.
For slickness, I wonder whether pivot_nested_intervals(..., nesting = "ph") (i.e., one level of nesting) should work to provide a nice wide dataframe with renamed columns.
Then maybe it should be pivot_textgrid_tiers(..., tiers) where if tiers is more than element, we handle nesting.
Hi! Thanks for the package, it's cool!
I wonder if you could implement a way to read TGs as "hierarchical", so that, given a hierarchical specification, you get for example:
Where the hierarchical structure is
sentence > word > phand each of them are a tier in the TG.