Postprocessing

audioclass.postprocess

Module for postprocessing audio classification model outputs.

This module provides functions to convert raw model outputs (class probabilities and features) into various formats suitable for analysis, visualization, or storage. The primary formats include:

• xarray Datasets: Structured datasets containing features, probabilities, and metadata like time, labels, location, and recording time.
• Lists of soundevent objects: Collections of PredictedTag and Feature objects, compatible with the soundevent library.

These functions facilitate seamless integration with downstream analysis tools and enable flexible representation of audio classification results.

Functions:

Name	Description
convert_to_dataset	Convert features and class probabilities to an xarray Dataset.
convert_to_features_array	Convert features to an xarray DataArray.
convert_to_features_list	Convert a feature array to a list of soundevent Feature objects.
convert_to_predicted_tags_list	Convert class probabilities to a list of predicted tags.
convert_to_probabilities_array	Convert class probabilities to a DataArray.

Attributes

Functions

convert_to_dataset(features, class_probs, labels, hop_size, start_time=0, latitude=None, longitude=None, recorded_on=None, attrs=None)

Convert features and class probabilities to an xarray Dataset.

Parameters:

Name	Type	Description	Default
features	ndarray	A 2D array of features, where each row corresponds to a frame and each column to a feature.	required
class_probs	ndarray	A 2D array of class probabilities, where each row corresponds to a frame and each column to a class.	required
labels	List[str]	A list of labels for the classes.	required
hop_size	float	The time step between frames in seconds.	required
start_time	float	The start time of the first frame in seconds. Defaults to 0.	0
latitude	Optional[float]	The latitude of the recording location. Defaults to None.	None
longitude	Optional[float]	The longitude of the recording location. Defaults to None.	None
recorded_on	Optional[datetime]	The date and time the recording was made. Defaults to None.	None
attrs	Optional[dict]	Additional attributes to add to the Dataset. Defaults to None.	None

Returns:

Type	Description
Dataset	An xarray Dataset containing the features and probabilities as DataArrays, along with coordinates and attributes.

convert_to_features_array(features, hop_size, start_time=0, latitude=None, longitude=None, recorded_on=None, attrs=None)

Convert features to an xarray DataArray.

Parameters:

Name	Type	Description	Default
features	ndarray	A 2D array of features, where each row corresponds to a frame and each column to a feature.	required
hop_size	float	The time step between frames in seconds.	required
start_time	float	The start time of the first frame in seconds. Defaults to 0.	0
latitude	Optional[float]	The latitude of the recording location. Defaults to None.	None
longitude	Optional[float]	The longitude of the recording location. Defaults to None.	None
recorded_on	Optional[datetime]	The date and time the recording was made. Defaults to None.	None
attrs	Optional[dict]	Additional attributes to add to the DataArray. Defaults to None.	None

Returns:

Type	Description
DataArray	An xarray DataArray with dimensions time and feature, containing the features.

convert_to_features_list(features, prefix)

Convert a feature array to a list of soundevent Feature objects.

Parameters:

Name	Type	Description	Default
features	ndarray	A 2D array of features, where each row corresponds to a frame and each column to a feature.	required
prefix	str	A prefix to add to each feature name.	required

Returns:

Type	Description
List[List[Feature]]	A list of lists of Feature objects, where each inner list corresponds to a frame and contains the features for that frame.

convert_to_predicted_tags_list(class_probs, tags, confidence_threshold=DEFAULT_THRESHOLD)

Convert class probabilities to a list of predicted tags.

Parameters:

Name	Type	Description	Default
class_probs	ndarray	A 2D array of class probabilities, where each row corresponds to a frame and each column to a class.	required
tags	List[Tag]	A list of Tag objects representing the possible classes.	required
confidence_threshold	float	The minimum probability threshold for a tag to be considered a prediction. Defaults to DEFAULT_THRESHOLD.	DEFAULT_THRESHOLD

Returns:

Type	Description
List[List[PredictedTag]]	A list of lists of PredictedTag objects, where each inner list corresponds to a frame and contains the predicted tags for that frame.

Raises:

Type	Description
ValueError	If the number of output tags does not match the number of columns in class_probs.

convert_to_probabilities_array(class_probs, labels, hop_size, start_time=0, latitude=None, longitude=None, recorded_on=None, attrs=None)

Convert class probabilities to a DataArray.

Parameters:

Name	Type	Description	Default
class_probs	ndarray	A 2D array of class probabilities, where each row corresponds to a frame and each column to a class.	required
labels	List[str]	A list of labels for the classes.	required
hop_size	float	The time step between frames in seconds.	required
start_time	float	The start time of the first frame in seconds. Defaults to 0.	0
latitude	Optional[float]	The latitude of the recording location. Defaults to None.	None
longitude	Optional[float]	The longitude of the recording location. Defaults to None.	None
recorded_on	Optional[datetime]	The date and time the recording was made. Defaults to None.	None
attrs	Optional[dict]	Additional attributes to add to the DataArray. Defaults to None.	None

Returns:

Type	Description
DataArray	An xarray DataArray with dimensions time and label, containing the class probabilities.