0:00:15so we break some mining or calibration schemes in an unconstrained environments
0:00:20facing fifty five combined foundation conditions of various duration and various noise snr types
0:00:27and reverse specially focusing on low as non short duration conditions
0:00:32general and calibrations teens employing quality
0:00:36do you can train conventionally on clean and four data so having three minutes of
0:00:41speech and no clean speech
0:00:44and use this on all conditions you facing but you will have a certain calibration
0:00:51calibration loss and what you could otherwise do is you can go for matched calibration
0:00:54so for each condition your training the colour we just a the calibration parameters but
0:01:00you will have a lot a high degree of freedom so double the amount of
0:01:03conditions you need to train as parameters and what we want to focus is
0:01:07on having a pasta models and no more amount of parameters to train so having
0:01:12low amount of parameters
0:01:14so what varying and male and in topic were proposing before was quality measure functions
0:01:19but use duration does not directly but that's not estimates and low snr are quite
0:01:26and we are using audio unified audio characteristics for quality vector estimate
0:01:32it was already proposed to use be linear kernel combination mattresses for this we of
0:01:36the reference quality vector and a probe quality vector
0:01:40but in here we have
0:01:41that's great amount of parameters of conditions as amount of parameters to train which is
0:01:46quite high and what we have opposing this function of quality estimates so to use
0:01:51the cousin of these quality vectors so we got to the degree of freedom of
0:01:54three witches
0:01:59and basically
0:02:02purpose so and here we depicted calibration mismatch
0:02:06from having the conventional scheme which is quite high and if we going for the
0:02:12matched calibration mismatch which is quite low we can approximate quite well as you posted