| 0:00:15 | so we break some mining or calibration schemes in an unconstrained environments |
|---|
| 0:00:20 | facing fifty five combined foundation conditions of various duration and various noise snr types |
|---|
| 0:00:27 | and reverse specially focusing on low as non short duration conditions |
|---|
| 0:00:32 | general and calibrations teens employing quality |
|---|
| 0:00:36 | do you can train conventionally on clean and four data so having three minutes of |
|---|
| 0:00:41 | speech and no clean speech |
|---|
| 0:00:44 | and use this on all conditions you facing but you will have a certain calibration |
|---|
| 0:00:49 | mismatch |
|---|
| 0:00:51 | calibration loss and what you could otherwise do is you can go for matched calibration |
|---|
| 0:00:54 | so for each condition your training the colour we just a the calibration parameters but |
|---|
| 0:01:00 | you will have a lot a high degree of freedom so double the amount of |
|---|
| 0:01:03 | conditions you need to train as parameters and what we want to focus is |
|---|
| 0:01:07 | on having a pasta models and no more amount of parameters to train so having |
|---|
| 0:01:12 | low amount of parameters |
|---|
| 0:01:14 | so what varying and male and in topic were proposing before was quality measure functions |
|---|
| 0:01:19 | but use duration does not directly but that's not estimates and low snr are quite |
|---|
| 0:01:25 | unstable |
|---|
| 0:01:26 | and we are using audio unified audio characteristics for quality vector estimate |
|---|
| 0:01:32 | it was already proposed to use be linear kernel combination mattresses for this we of |
|---|
| 0:01:36 | the reference quality vector and a probe quality vector |
|---|
| 0:01:40 | but in here we have |
|---|
| 0:01:41 | that's great amount of parameters of conditions as amount of parameters to train which is |
|---|
| 0:01:46 | quite high and what we have opposing this function of quality estimates so to use |
|---|
| 0:01:51 | the cousin of these quality vectors so we got to the degree of freedom of |
|---|
| 0:01:54 | three witches |
|---|
| 0:01:56 | karen |
|---|
| 0:01:58 | arguable |
|---|
| 0:01:59 | and basically |
|---|
| 0:02:02 | purpose so and here we depicted calibration mismatch |
|---|
| 0:02:06 | from having the conventional scheme which is quite high and if we going for the |
|---|
| 0:02:12 | matched calibration mismatch which is quite low we can approximate quite well as you posted |
|---|