Is the standard WFST approach implemented on GPU as well as CPU?
I know that the neural net approach uses the GPU by default, but to implement the WFST approach on a GPU I should use the CUDA Matrix library. Is this correct?
Thanks,
- Jeff
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Is the standard WFST approach implemented on GPU as well as CPU?
I know that the neural net approach uses the GPU by default, but to
implement the WFST approach on a GPU I should use the CUDA Matrix library.
Is this correct?
The entire decoding process. I'm looking to find out what speech recognition packages currently utilize a GPU for recognition. I saw the wiki page about the CUDA matrix library and wasn't sure if that applied to the non-neural network approach (what I had called the standard WFST approach). Is the library optional or required?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
The entire decoding process. I'm looking to find out what speech
recognition packages currently utilize a GPU for recognition. I saw the wiki
page http://kaldi.sourceforge.net/cudamatrix.html about the CUDA matrix
library and wasn't sure if that applied to the non-neural network approach
(what I had called the standard WFST approach). Is the library optional or
required?
Hi,
Is the standard WFST approach implemented on GPU as well as CPU?
I know that the neural net approach uses the GPU by default, but to implement the WFST approach on a GPU I should use the CUDA Matrix library. Is this correct?
Thanks,
- Jeff
By WFST approach you mean the Gaussian computations or the entire decoding
process?
Or some other WFST operation?
Paul
On 2 April 2014 21:35, Jeff Acquaviva jacquaviva@users.sf.net wrote:
The entire decoding process. I'm looking to find out what speech recognition packages currently utilize a GPU for recognition. I saw the wiki page about the CUDA matrix library and wasn't sure if that applied to the non-neural network approach (what I had called the standard WFST approach). Is the library optional or required?
ATM, the GPU is mainly used for neural networks training and not much else.
y.
On Wed, Apr 2, 2014 at 4:15 PM, Jeff Acquaviva jacquaviva@users.sf.netwrote: