Smoothed Cepstral Peak Prominence (CPPS) in one call

Computes the smoothed cepstral peak prominence (CPPS, in dB) of a `Sound` — a widely used acoustic measure of voice quality/breathiness. It performs the whole PowerCepstrogram-then-CPPS pipeline in a single call and returns the same value as `sound$to_powercepstrogram(...)$get_cpps(...)` with matching parameters. You supply the analysis parameters directly and are responsible for passing valid values (see the arguments below).

Usage

calculate_cpps_fast(
  sound,
  subtract_tilt = TRUE,
  time_averaging_window = 0.001,
  quefrency_averaging_window = 5e-04,
  pitch_floor = 60,
  pitch_ceiling = 333.3,
  delta_f0 = 0.05,
  interpolation = "parabolic",
  qstart_fit = 0.003,
  qend_fit = 0.04,
  trend_line_type = "straight",
  fit_method = "robust",
  cepstrogram_pitch_floor = 60,
  time_step = 0.002,
  max_frequency = 5000,
  pre_emphasis_from = 50
)

Arguments

sound

Sound object or external pointer

subtract_tilt

Logical, subtract tilt before calculating CPPS (default TRUE)

time_averaging_window

Numeric, time averaging window in seconds (default 0.001)

quefrency_averaging_window

Numeric, quefrency averaging window in seconds (default 0.0005)

pitch_floor

Numeric, minimum F0 in Hz (default 60)

pitch_ceiling

Numeric, maximum F0 in Hz (default 333.3)

delta_f0

Numeric, F0 fractional precision (default 0.05)

interpolation

Character, one of "parabolic", "none", "cubic", "sinc70", "sinc700" (default "parabolic")

qstart_fit

Numeric, quefrency range start for fitting in seconds (default 0.003)

qend_fit

Numeric, quefrency range end in seconds (default 0.04)

trend_line_type

Character, "straight" or "exponential" (default "straight")

fit_method

Character, "robust" (Siegel repeated median), "least_squares", or "robust slow" (Theil-Sen). Default "robust". **"robust slow" is not reproducible**: it samples randomly inside Praat's slope selection, so repeated runs on the same input differ (~0.8 dB observed) and can return values on the order of 1e290. That is an upstream Praat defect, reproduced faithfully here; pladdrr warns once per session when you select it.

cepstrogram_pitch_floor

Numeric, pitch floor for cepstrogram creation (default 60)

time_step

Numeric, time step for cepstrogram in seconds (default 0.002)

max_frequency

Numeric, max frequency for cepstrogram in Hz (default 5000)

pre_emphasis_from

Numeric, pre-emphasis frequency in Hz (default 50)

Value

A single numeric CPPS value in dB.

Details

Use this when analysing many files in a loop. For interactive or one-off use, the object API (`sound$to_powercepstrogram(...)$get_cpps(...)`) is equivalent and validates its inputs more forgivingly; this function skips that validation, so pass well-formed parameters.

This is a **CPPS** helper: it builds a whole-sound PowerCepstrogram and then computes a smoothed peak-prominence summary. If you need a single-interval **CPP** value that matches Praat's `To PowerCepstrum` workflow, extract the interval, convert it to a Spectrum, and call `to_power_cepstrum()$get_peak_prominence()` instead. That path is much cheaper and is not the same metric.

Defaults differ from Praat's

These defaults follow the AVQI/clinical convention, **not** the defaults of Praat's `PowerCepstrogram: Get CPPS...` dialog. On a 1 s test signal the two parameter sets give 9.92 dB and 4.82 dB respectively — a different measurement, not a rounding difference. Pass the Praat values explicitly to reproduce a Praat run:

| parameter | Praat default | pladdrr default | |—|—|—| | `time_averaging_window` | 0.02 | 0.001 | | `quefrency_averaging_window` | 0.0005 | 0.0005 | | `pitch_floor` | 60 | 60 | | `pitch_ceiling` | 330 | 333.3 | | `qstart_fit` | 0.001 | 0.003 | | `qend_fit` | 0.05 | 0.04 | | `trend_line_type` | `"exponential"` | `"straight"` | | `fit_method` | `"robust slow"` | `"robust"` |

The `fit_method` difference is deliberate beyond convention: Praat's `"robust slow"` (Theil-Sen) is not reproducible — see the `fit_method` argument.

Examples

sound <- Sound(system.file("extdata", "test.wav", package = "pladdrr"))

# Full-file CPPS calls on the 60 s demo file are CPU-heavy (> 5 s under
# R CMD check), so they live in \donttest{} (still run under
# --run-donttest).
# \donttest{
# One-call CPPS with the default (AVQI-convention) parameters
calculate_cpps_fast(sound)
#> [1] 9.920529

# The object API gives the identical value
pcep <- sound$to_powercepstrogram(60, 0.002, 5000, 50)
pcep$get_cpps()
#> [1] 9.920529

# Reproduce Praat's own "Get CPPS..." defaults
calculate_cpps_fast(sound,
  time_averaging_window = 0.02, pitch_ceiling = 330,
  qstart_fit = 0.001, qend_fit = 0.05,
  trend_line_type = "exponential", fit_method = "least_squares"
)
#> [1] 5.671056
# }
# Single-interval CPP is a different, cheaper path
segment <- sound$extract_part(0, 0.5)
cpp <- segment$to_spectrum()$to_power_cepstrum()$get_peak_prominence(
  60, 333.3, "parabolic", 0.001, 0.05, "exponential decay", "robust slow"
)