rtvamp.FeatureComputation

class rtvamp.FeatureComputation(samplerate: float)

Compute features with (multiple) plugins block-wise or from a stream (advanced usage).

__init__(samplerate: float)

Initialize FeatureComputation class.

Parameters:

samplerate – Input sample rate

Methods

__init__(samplerate)	Initialize FeatureComputation class.
add_plugin(key[, parameter, paths])	Add plugin for processing.
get_output_descriptors()	Get output descriptors.
initialise(blocksize[, stepsize])	Initialise all added plugins.
process_block(timedata_block, timestamp)	Process a single block/frame of data.
process_signal(timedata[, timestamp_start])	Process data of arbitrary length.
reset()	Reset all added plugins.

Attributes

outputs	List of all plugin outputs.
plugins	List of added plugins.

property plugins: list[Plugin]

List of added plugins.

property outputs: list[str]

List of all plugin outputs.

add_plugin(key: str, parameter: dict[str, float] | None = None, paths: list[PathLike] | None = None)

Add plugin for processing.

Parameters:

• key – Plugin key/identifer as returned by e.g. list_plugins()

• parameter – Dict with parameter identifiers and values. Use get_plugin_metadata() or Plugin.get_parameter_descriptors() to list available parameters and their constraints.

• paths – Custom paths, either search paths or plugin library paths

initialise(blocksize: int, stepsize: int | None = None)

Initialise all added plugins.

Parameters:

• blocksize – Block size in samples

• stepsize – Step size in samples (< blocksize, default = blocksize)

reset()

Reset all added plugins.

get_output_descriptors()

Get output descriptors.

process_block(timedata_block: ndarray, timestamp: float) → List[List[float]]

Process a single block/frame of data.

Parameters:

• timedata_block – Single block/frame of time series data (length equal to initialised blocksize)

• timestamp – Timestamp of block in seconds

Returns:

List of computed features (same length as outputs).

The feature itself is list of floats. Check bin_count with get_output_descriptors().

process_signal(timedata: ndarray, timestamp_start: float = 0) → tuple[ndarray, list[ndarray]]

Process data of arbitrary length.

Parameters:

• timedata – Time series data of arbitrary length. Signal will be cropped to blocks accoring to initialised stepsize and blocksize.

• timestamp_start – Timestamp of signal start in seconds

Returns:

• Array of timestamps in seconds

• List of arrays of computed features (same length as timestamps). Check outputs to map the arrays to the plugin outputs.