pumpp.feature.HCQT

class pumpp.feature.HCQT(name, sr, hop_length, n_octaves=8, over_sample=3, fmin=None, harmonics=None, log=False, conv='channels_last')[source]

Harmonic Constant-Q transform

Attributes:
name : str

The name for this feature extractor

sr : number > 0

The sampling rate of audio

hop_length : int > 0

The number of samples between CQT frames

n_octaves : int > 0

The number of octaves in the CQT

over_sample : int > 0

The amount of frequency oversampling (bins per semitone)

fmin : float > 0

The minimum frequency of the CQT

harmonics : list of int >= 1

The list of harmonics to compute

log : boolean

If True, scale the magnitude to decibels

Otherwise, use linear magnitude

conv : {‘tf’, ‘th’, ‘channels_last’, ‘channels_first’, None}

convolution dimension ordering:

  • ‘channels_last’ for tensorflow-style 2D convolution
  • ‘tf’ equivalent to ‘channels_last’
  • ‘channels_first’ for theano-style 2D convolution
  • ‘th’ equivalent to ‘channels_first’
__init__(name, sr, hop_length, n_octaves=8, over_sample=3, fmin=None, harmonics=None, log=False, conv='channels_last')[source]

Initialize self. See help(type(self)) for accurate signature.

Methods

__init__(name, sr, hop_length[, n_octaves, …]) Initialize self.
layers() Construct Keras input layers for the given transformer
merge(data) Merge an array of output dictionaries into a single dictionary with properly scoped names.
n_frames(duration) Get the number of frames for a given duration
phase_diff(phase) Compute the phase differential along a given axis
pop(field)
register(key, dimension, dtype[, channels]) Register a field as a tensor with specified shape and type.
scope(key) Apply the name scope to a key
transform(y, sr) Transform an audio signal
transform_audio(y) Compute the HCQT

Attributes

idx