Binary onset and offset maps (transientMapProc.m
)¶
The information about sudden intensity changes, as represented by onsets or
offsets, can be combined in order to organise and group the acoustic input
according to individual auditory events. The required processing is similar for
both onsets and offsets, and is summarised by the term transient detection. To
apply this transient detection based on the onset strength or offset strength,
the user should use the request name ’onset_map’
or ’offset_map’
,
respectively. Based on the transient strength which is derived from the
corresponding onset strength and offset strength processor (described in
Onset strength (onsetProc.m) and Offset strength (offsetProc.m), a binary decision
about transient activity is formed, where only the most salient information is
retained. To achieve this, temporal and across-frequency constraints are imposed
for the transient information. Motivated by the observation that two sounds are
perceived as separated auditory events when the difference in terms of their
onset time is in the range of 20 ms – 40 ms [Turgeon2002], transients are fused
if they appear within a pre-defined time context. If two transients appear
within this time context, only the stronger one will be considered. This time
context can be adjusted by trm_fuseWithinSec
. Moreover, the minimum
across-frequency context can be controlled by the parameters trm_minSpread
.
To allow for this selection, individual transients which are connected across
multiple TF units are extracted using Matlab’s image labelling tool
bwlabel
. The binary transient map will only retain those transients which
consists of at least trm_minSpread
connected TF units. The salience of the
cue can be specified by the detection thresholds trm_minStrengthdB
. Whereas
this thresholds control the required relative change, a global threshold
excludes transient activity if the corresponding rate-map level is below a
pre-defined threshold, as determined by trm_minValuedB
. A summary of all
parameters is given in Table 28.
Parameter | Default | Description |
---|---|---|
trm_fuseWithinSec |
30E-3 |
Time constant below which transients are fused |
trm_minSpread |
5 |
Minimum number of connected TF units |
trm_minStrengthdB |
3 |
Minimum onset strength in dB |
trm_minValuedB |
-80 |
Minimum rate-map level in dB |
To illustrate the benefit of selecting onset and offset information, a rate-map
representation is shown in Fig. 30 (left panel), where the
corresponding onsets and offsets detected by the transientMapProc
, through
two individual requests ’onset_map’
and ’offset_map’
, and without
applying any temporal or across-frequency constraints are overlaid (respectively
in black and white). It can be seen that the onset and offset information is
quite noisy. When only retaining the most salient onsets and offsets by applying
temporal and across-frequency constraints (right panel), the remaining onsets
and offsets can be used as temporal markers, which clearly mark the beginning
and the end of individual auditory events.
[Turgeon2002] | Turgeon, M., Bregman, A. S., and Ahad, P. A. (2002), “Rhythmic masking release: Contribution of cues for perceptual organization to the cross-spectral fusion of concurrent narrow-band noises,” Journal of the Acoustical Society of America 111(4), pp. 1819–1831. |