Flag tokens whose f0 has gaps inside the measured region

Adds two token-level columns so unvoiced stretches are visible in the exported data rather than being silently scattered NAs: n_missing, the number of that token's rows with no f0, and has_gap, TRUE when at least one of those rows sits between two rows that do have f0.

Usage

flag_f0_gaps(df, token = "token", f0 = "f0", time = "time")

Arguments

df

Long-format f0 data frame.

token, f0, time

Column names. Defaults "token", "f0", "time".

Value

df with the n_missing and has_gap columns added.

Details

The distinction matters because a missing value at the edge of a region usually just means the region ran slightly past the voicing, while a missing value in the middle means voicing was interrupted (a medial voiceless stop, a creaky patch, or a tracking failure). Only the latter sets has_gap, so has_gap marks the tokens worth looking at in F0 Correction. Run this on the native frames rather than on a resampled grid: resampling can fill a short dropout from its nearer frame, leaving no NA behind for this to find.

A token with no f0 at all gets has_gap = FALSE: there is no interior to speak of. Existing columns of either name are replaced.

See also

resample_f0_equal(), trim_to_voiced().