dcase_framework.recognizers.EventRecognizer¶

class dcase_framework.recognizers.EventRecognizer(*args, **kwargs)[source]¶

Multi-class multi-label detection

Parameters

Field name	Value type	Description
frame_accumulation, Defining frame probability accumulation.
enable	bool	Enable frame probability accumulation.
type	string {sliding_sum \| sliding_mean \| sliding_median }	Operator type used to accumulate.
window_length_seconds	float	Window length in seconds for sliding accumulation.
frame_binarization, Defining frame probability binarization.
enable	bool	Enable frame probability binarization.
type	string {frame_max \| global_threshold }	Type of binarization: `frame_max`, each frame is treated individually, max of each frame is set to one, all others to zero. `global_threshold`, global threshold, all values over the threshold are set to one.
threshold	float	Threshold value. Set to null if not used.
event_activity_processing, Event activity processing per frame.
enable	bool	Enable activity processing.
type	string {median_filtering}	Type of decision: `median_filtering`, median filtering of decision inside window.
window_length_seconds	float	Length of sliding window in seconds for activity processing.
event_post_processing, Event post processing per event.
enable	bool	Enable event processing.
minimum_event_length_seconds	float	Minimum allowed event length. Shorter events will be removed.
minimum_event_gap_second	float	Minimum allowed gap between events. Smaller gaps between events will cause events to be merged together.

Constructor

Parameters:

hop_length_seconds : float

Analysis frame hop length in seconds. Default value 0.02

params : dict

Processing parameters

class_labels : list of str

Class labels in a list

__init__(*args, **kwargs)[source]¶