Duration of each f0 contour

Adds the time from the first to the last frame with a measured f0 in each unit: a whole token, or each labelled segment of a landmark set (e.g. each syllable) within a token. This is the span the contour covers, not the length of the unit's interval: unmeasured frames at either edge do not count, and unmeasured frames inside the span (a gap) do not shorten it.

Usage

contour_duration(
  df,
  token = "token",
  time = "time",
  f0 = "f0",
  set = NULL,
  name = NULL,
  min_run = 2
)

Arguments

df

Long-format f0 data frame.

token, time, f0

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

set

Landmark-set base name (e.g. "syllable") to measure each segment of that set, from its <set>_start / <set>_end columns (see attach_landmarks()). NULL (default) measures whole tokens.

name

Name of the new column. Default "token_f0_dur", or "<set>_f0_dur" with set.

min_run

Minimum number of consecutive measured frames the first and last frame must belong to. Default 2; 1 takes the first and last measured frames as they are.

Value

df with the duration column appended, in the units of the time column. Row order is preserved.

Details

A frame counts as measured when its f0 is a number other than 0; NA and 0 (which pitch trackers write for unvoiced frames) do not count. Negative values do, so a normalised f0 column (semitones, z-scores) works as well as Hz. A unit with a single measured frame gets 0, and a unit with none gets NA.

The first and last frames are anchored on a run of at least min_run consecutive measured frames, the rule trim_to_voiced() uses for the voiced region. A lone measured frame cut off from the contour by unmeasured frames, such as a tracking artefact in the silence, then cannot stretch the duration. If no run is that long, the first and last measured frames are used. On native frames the default makes a whole token's duration equal to its voiced span (voiced_s in the F0 Extraction summary). Input without unmeasured rows (a .PitchTier, or an export with those rows dropped) has no gaps to see, so there every measured frame counts.

With set, frames are grouped by token and by the segment's _start / _end boundaries. Only labelled segments are measured: frames in an empty-label interval (silence, as written by attach_landmarks()) or outside the tier get NA.

The value is repeated on every row of its unit, so the column can serve as a token- or segment-level predictor directly.

See also

trim_to_voiced() for the voiced span as a row filter; normalise_time_landmarks() for a time axis within the same segments.

Examples

df <- data.frame(token = "a", time = seq(0, 0.5, by = 0.1),
                 f0 = c(NA, 200, 210, NA, 190, NA))
contour_duration(df)$token_f0_dur   # 0.4 - 0.1 = 0.3, on every row
#> [1] 0.1 0.1 0.1 0.1 0.1 0.1