[keyonjie]

This is submitted to fix the issue #3112

Refine to use the consistent activation threshold for different formats,
use a unified divider 0.05, and update the threshold at .params() stage.

These refinement will fix the KWD trigger volume inconsistent for
different sample format issue.

Signed-off-by: Keyon Jie yang.jie@linux.intel.com

[ClarexZhou]

verified, issue #3112 can be fixed by this PR.

[lgirdwood]

@singalsu do these value make sense for given types and gain ?

[singalsu]

@keyonjie @lgirdwood Yes, the thresholds need to be scaled for their PCM format. But those multiplier values are not dividers, I'd call 0.05 as ACTIVATION_DEFAULT_THRESHOLD instead. Also since it's a float it could be converted this way. But it's a matter of taste what you feel is the best.

#define ACTIVATION_DEFAULT_THRESHOLD_S32 Q_CONVERT_FLOAT(ACTIVATION_DEFAULT_THRESHOLD, 31) /* Q1.31 */
#define ACTIVATION_DEFAULT_THRESHOLD_S24 Q_CONVERT_FLOAT(ACTIVATION_DEFAULT_THRESHOLD, 23) /* Q1.23 */
#define ACTIVATION_DEFAULT_THRESHOLD_S16 Q_CONVERT_FLOAT(ACTIVATION_DEFAULT_THRESHOLD, 15) /* Q1.15 */

[keyonjie]

Yes @singalsu the naming 'divider' here is somewhat, it should be something like coefficient or multiplier for calculating the threshold. But since it is fixing issue here I would prefer to keep the historic naming and leave the float converting to future refinement, for better bisect also.

[lgirdwood]

@keyonjie I'd prefer we only change this once as threshold will change slightly everytime we make this change so lets get it right fist time. Please align with @singalsu method (which is using ints and not floats in the runtime code - i.e. float is converted to int by the macros above).

[keyonjie]

@keyonjie I'd prefer we only change this once as threshold will change slightly everytime we make this change so lets get it right fist time. Please align with @singalsu method (which is using ints and not floats in the runtime code - i.e. float is converted to int by the macros above).

yes I can convert to use those macros to make it unified, though actually both method are using ints and not floats in the runtime code, float is already converted to int with today's code.