Train noise data to let HMM more robust

[stevenyslin]

Hello,

In my case, I train my voice in quiet environment and have a good accuracy,
but when test in noise environment, the accuracy will be very bad,
so I want to train mixed voice (quiet data + noise data) to let my model more robust.

My question：
Q1：
Could I add mixed voice to train my own HMM model?

Q2：
Could I add all of list in my my_db.filler, or just add some in my case?

++AH++               +AH+
++BEEP++             +TONE+
++BREATH++           +BREATH+
++CHAIR_SQUEAK++     +SQUEAK+
++CLICK++            +CLICK+
++DISK_NOISE++       +NOISE+
++DOOR_SLAM++        +SLAM+
++EXHALATION++       +BREATH+
++GRUNT++            +GRUNT+
++INHALATION++       +BREATH+
++LAUGHTER++         +LAUGH+
++LIP_SMACK++        +SMACK+
++LOUD_BREATH++      +BREATH+
++MICROPHONE_MVT++   +NOISE+
++MISC_NOISE++       +NOISE+
++MOVEMENT++         +NOISE+
++NOISE++            +NOISE+
++PAPER_RUSTLE++     +NOISE+
++PHONE_RING++       +RING+
++SIGH++             +SIGH+
++TAP++              +TAP+
++THROAT_CLEAR++     +GRUNT+
++TONGUE_CLICK++     +CLICK+
++TYPING++           +TYPING+
<sil>   SIL
</s>    SIL
<s> SIL

Thanks for your help.

[Nickolay V. Shmyrev]

Could I add mixed voice to train my own HMM model?

It is not clear what are you asking.

Could I add all of list in my my_db.filler, or just add some in my case?

You could but it does not have much practical sense.