MFCC

Mel-frequency cepstral coefficients for speech and speaker recognition.

Arguments

.xptr

Not for direct use. External pointer to the underlying C++ MFCC object; set internally when a method returns a new MFCC.

Value

An MFCC object with methods for Mel-frequency cepstral coefficient analysis.

Details

MFCCs are widely used features in speech and speaker recognition systems. They represent the short-term power spectrum of a sound on a mel scale, which approximates human auditory perception. Uses a shared dispatch table for minimal memory per object.

Creating MFCC objects

• sound$to_mfcc() - extract MFCCs with default parameters

Query methods

• get_number_of_frames() - number of analysis frames

• get_time_step() - time step between frames

• get_max_num_coefficients() - maximum number of coefficients

• get_fmin(), get_fmax() - frequency range (mel)

• get_c0_at_frame(frame) - C0 (energy) for a specific frame

• get_value_in_frame(frame, coef) - coefficient value at a frame

• get_coefficients_at_frame(frame) - all coefficients for a frame

• get_all_coefficients() - matrix of all coefficients

• get_all_c0() - vector of all C0 values

Liftering

• lifter(L) - apply cepstral liftering (weighting)

Export

• as_data_frame(include_c0) - convert to a data.frame/data.table

• to_matrix() - convert to a Matrix object

See also

Sound, MelSpectrogram

Examples

sound <- Sound$create_tone(frequency = 150, duration = 0.3)
mfcc <- sound$to_mfcc(
  num_coefficients = 13,
  analysis_width = 0.025,
  time_step = 0.01,
  f1_mel = 100,
  fmax_mel = 7800,
  df_mel = 100
)
n_frames <- mfcc$get_number_of_frames()
coefs <- mfcc$get_all_coefficients()
mfcc$lifter(22)
df <- mfcc$as_data_frame(include_c0 = TRUE)