Add event combination post-processors based on inter-event gaps
The current event combination filters combine events from the baseline (start, or low frequency) of the first event.
The idea is to anchor a composite event to the first detected component because that component tends to be the strongest and most defining event.
However, these combination filters do not work when the goal is to chain a series of events together. For a chaining operation the inter-event gap should be used as the threshold for joining events (rather than the baseline).
Proposal:
Add a CombineEventSequence post processing step with the following parameters:
Interval? TemporalGap
Interval? SpectralGap
Interval? SequenceLength
Interval? TemporalDifference
Interval? SpectralDifference
Alternatives:
We could split the proposal up into two separate filters - but I want to explore one filter that could work for events on a diagonal axis.
Additional context
Several of the NSW DPI recognisers could benefit from this change.
This should be done after #468.
The existing documentation for the SyllableSequence and CombineVerticalSyllables post processing steps should be updated
Add event combination post-processors based on inter-event gaps
The current event combination filters combine events from the baseline (start, or low frequency) of the first event.
The idea is to anchor a composite event to the first detected component because that component tends to be the strongest and most defining event.
However, these combination filters do not work when the goal is to chain a series of events together. For a chaining operation the inter-event gap should be used as the threshold for joining events (rather than the baseline).
Proposal:
Add a
CombineEventSequencepost processing step with the following parameters:Alternatives:
We could split the proposal up into two separate filters - but I want to explore one filter that could work for events on a diagonal axis.
Additional context
Several of the NSW DPI recognisers could benefit from this change.
This should be done after #468.
The existing documentation for the
SyllableSequenceandCombineVerticalSyllablespost processing steps should be updated