High WER for system combination(SGMM+DNN)

[Bhargav Pulugundla]

Hi all,

What could be the reason for the WER with system combination(SGMM+DNN) to be much higher than DNN and SGMM individually?

%WER (SGMM+DNN) = 52.49 [ 21573 / 41103, 245 ins, 17703 del, 3625 sub ]
%WER(DNN) = 24.10 [ 9905 / 41103, 1527 ins, 2766 del, 5612 sub ]
%WER(SGMM) = 20.94 [ 8606 / 41103, 1837 ins, 2017 del, 4752 sub ]

Thanks.

[Paul Dixon]

How are you combing the systems?

On 3 November 2014 17:29, Bhargav Pulugundla bpulugundla@users.sf.net
wrote:

Hi all,

What could be the reason for the WER with system combination(SGMM+DNN) to
be much higher than DNN and SGMM individually?

%WER (SGMM+DNN) = 52.49 [ 21573 / 41103, 245 ins, 17703 del, 3625 sub ]
%WER(DNN) = 24.10 [ 9905 / 41103, 1527 ins, 2766 del, 5612 sub ]
%WER(SGMM) = 20.94 [ 8606 / 41103, 1837 ins, 2017 del, 4752 sub ]

Thanks.

High WER for system combination(SGMM+DNN)
https://sourceforge.net/p/kaldi/discussion/1355348/thread/24229f5f/?limit=25#497f

---

Sent from sourceforge.net because you indicated interest in
https://sourceforge.net/p/kaldi/discussion/1355348/

To unsubscribe from further messages, please visit
https://sourceforge.net/auth/subscriptions/

[Bhargav Pulugundla]

The error was in my phoneset mapping. I rectified it and now the WER is 20.50. Thanks.

By the way, I was combining the lattices(through the script score_combine.) to combine the systems.