sample.sample module

Module for using the entire SAMPLE method pipeline

class sample.sample.SAMPLE(sinusoidal: ~sample.sms.sm.SinusoidalModel = None, regressor: ~sample.hinge.HingeRegression = None, regressor_k: str = 'k_', regressor_q: str = 'q_', freq_reduce: ~typing.Callable[[~numpy.ndarray], float] = <function mean>, max_n_modes: ~typing.Optional[int] = None, **kwargs)

Bases: RegressorMixin, BaseEstimator

SAMPLE (Spectral Analysis for Modal Parameter Linear Estimate) model

Parameters:

• sinusoidal (SinusoidalModel) – Sinusoidal analysis model. Default is an instance of sample.sms.mm.ModalModel

• regressor – Modal parameters regression model. Default is an instance of sample.hinge.HingeRegression

• regressor_k (str) – Attribute name for the estimated slope coefficient of regressor

• regressor_q (str) – Attribute name for the estimated intercept coefficient of regressor

• freq_reduce (callable) – Callable function for reducing the frequency track into a single frequency. Defaults to numpy.mean()

• max_n_modes (int) – Number of modes to use in resynthesis. If None (default), then synthesise all modes

• **kwargs – Additional parameters for sub-models. See sample.sms.mm.ModalTracker, sample.sms.mm.ModalModel, sample.hinge.HingeRegression, and sample.utils.learn.OptionalStorage

param_matrix_

3-by-N matrix of modal parameters

Type:

array

property amps_: ndarray

Learned modal amplitudes

property decays_: ndarray

Learned modal decays

property energies_: ndarray

Learned modal energies

fit(x: ndarray, y=None, **kwargs)

Analyze audio data

Parameters:

• x (array) – audio input

• y (ignored) – exists for compatibility

• kwargs – Any parameter, overrides initialization

Returns:

self

Return type:

SAMPLE

property freqs_: ndarray

Learned modal frequencies

mode_argsort_(order: str = 'energies', reverse: bool = True) → List[int]

Get the indices for sorting modes

Parameters:

• order (str) – Feature to use for ordering modes. Default is "energies", so that reducing active modes keeps the modes with most energy. Other options are "freqs", "amps" and "decays"

• reverse (bool) – Whether the order should be reversed (decreasing). Defaults to True

Returns:

Array of indices for sorting

Return type:

array

predict(x: ndarray, n_modes: Optional[int] = None, order: str = 'energies', reverse: bool = True, **kwargs) → ndarray

Resynthesize audio

Parameters:

• x (array) – Time axis

• n_modes (int) – Number of modes to synthesize. If None, then use the max_n_modes parameter

• order (str) – Feature to use for ordering modes. Default is "energies", so that reducing active modes keeps the modes with most energy. Other options are "freqs", "amps" and "decays"

• reverse (bool) – Whether the order should be reversed (decreasing). Defaults to True

• **kwargs – Keyword arguments for additive_synth()

Returns:

Array of audio samples

Return type:

array

property regressor

Modal parameters regression model

sdt_params_(order: str = 'energies', reverse: bool = True) → dict

SDT parameters as a JSON serializable dictionary

Parameters:

• order (str) – Feature to use for ordering modes. Default is "energies", so that reducing active modes in SDT keeps the modes with most energy. Other options are "freqs", "amps" and "decays"

• reverse (bool) – Whether the order should be reversed (decreasing). Defaults to True

Returns:

SDT parameters

Return type:

dict

set_params(**kwargs)

Set the parameters of this estimator.

The method works on simple estimators as well as on nested objects (such as Pipeline). The latter have parameters of the form <component>__<parameter> so that it’s possible to update each component of a nested object.

Parameters:

**params (dict) – Estimator parameters.

Returns:

self – Estimator instance.

Return type:

estimator instance

property sinusoidal

Sinusoidal analysis model

sample.sample.additive_synth(x, freqs: Sequence[float], decays: Sequence[float], amps: Sequence[float], phases: Optional[Union[Sequence[float], str]] = None, analytic: bool = False, **kwargs) → array

Additively synthesize audio

Parameters:

• x (array) – Time axis

• freqs (array) – Modal frequencies

• decays (array) – Modal decays

• amps (array) – Modal amplitudes

• phases (array) – Starting phase for every mode, optional

• analytic (bool) – If True, use a complex exponential as an oscillator

• **kwargs – Keyword arguments for random phase generator

Returns:

Array of audio samples

Return type:

array

sample.sample.modal_energy(a: ndarray, d: ndarray, out: Optional[ndarray] = None) → ndarray

Compute the integrals of the modal amplitude envelopes

Parameters:

• a (array) – Amplitudes

• d (array) – Decays

• out (array) – Optional. Array to use for storing results

Returns:

Energies

Return type:

array