PowerCepstrum

A PowerCepstrum is the power spectrum of the log power spectrum, a representation that separates the source (glottal pulse, low quefrency) from the filter (vocal tract, high quefrency). Created from a Spectrum or extracted from a PowerCepstrogram at a specific time. The primary voice quality metric from this object is CPP (Cepstral Peak Prominence).

Value

A PowerCepstrum object.

Query methods

get_qmin(), get_qmax() - quefrency range (s)
get_quefrency_range() - quefrency range as a vector
get_n_bins() - number of quefrency bins
get_dq() - quefrency step (s)
get_q1() - starting quefrency value (s)

Peak analysis

get_peak_prominence(pitch_floor, pitch_ceiling, ...) - CPP value (dB), the main voice quality metric
get_peak_prominence_hillenbrand(pitch_floor, pitch_ceiling) - CPP using the Hillenbrand algorithm
get_quefrency_of_peak(interpolation) - quefrency of the cepstral peak (s)
get_value_at_quefrency(quefrency, interpolation, unit) - cepstral amplitude at a quefrency

Trend and smoothing

smooth(averaging_window) - smooth the cepstrum
fit_trend_line(qmin, qmax, trend_type, fit_method) - fit a regression trend line
get_trend_line_value(quefrency, ...) - value of the fitted trend at a quefrency
subtract_trend(qstart_fit, qend_fit, ...) - subtract the regression trend (returns a new PowerCepstrum)
subtract_trend_inplace(qstart_fit, qend_fit, ...) - subtract the trend in place (mutates)

Export and transform

as_matrix(), as_data_frame() - export
to_spectrum(random_phases) - convert back to Spectrum
to_matrix() - export as a matrix

Examples

sound <- Sound$create_tone(duration = 0.5, frequency = 200, sampling_rate =
 44100)
spectrum <- sound$to_spectrum()
cepstrum <- spectrum$to_power_cepstrum()
cpp <- cepstrum$get_peak_prominence()

# The same analysis on a recording read from disk
wav <- system.file("extdata", "test.wav", package = "pladdrr")
cepstrum <- Sound(wav)$to_spectrum()$to_power_cepstrum()