feat.parameter

[Suong-Sun Hong]

Hi,

I'm having trouble changing the window length for the sphinx.
If I change wlen in the feat.parameter file, it always go back to wlen =
0.0256.
Is there a way to change this value?

Thanks

[Nickolay V. Shmyrev]

If I change wlen in the feat.parameter file, it always go back to wlen =
0.0256.

You need to be accurate which file to change. The file is called
feat.params and it needs to be placed in the model folder. Most likely
some of the conditions aren't met. Because it works as expected here.

[Suong-Sun Hong]

Hi,

It still keeps going back to wlen=0.0256.
I'm modifying feat.param in the etc folder.
I have also tried to change the setting in the pocketsphinx-.0.6.1 model
folder.
However, I still face the same problem.
Where do you mean by 'model' folder?

Also by changing wlen value, I'm trying to adapt the speaking rate (frame
rate).
What other parameters in feat.param file can I change to adapt the speaking
rate?

Thanks

[Nickolay V. Shmyrev]

'm modifying feat.param in the etc folder.

What etc folder are you talking about? Did you train your own model? In that
case feat.param from etc is not used. You need to modify feat.param in model
folder. Model is located in

<training_folder>/model_parameters/model_name.cd_cont_<numberofsenones>

I'm trying to adapt the speaking rate (frame rate). What other parameters in
feat.param file can I change to adapt the speaking rate?

Frame rate is unrelated to speaking rate adaptation, it only sets the way to
read the audio. There are no params that control speaking rate or to adapt to
speaking rate accuracy, only acoustic model adaptation and dictionary
adaptation make sense if you want to recognize fast speech with good accuracy.
Dictionary adaptation is even more important because with fast speech the
phonetic presentation of the word changes drastically.

[Suong-Sun Hong]

so basically, I'm training my own acoustic model following the instruction at:
http://cmusphinx.sourceforge.net/wiki/tutorialam
And etc folder is created when I run:
../SphinxTrain/scripts_pl/setup_SphinxTrain.pl -task training model
../pocketsphinx/scripts/setup_sphinx.pl -task trainingmodel

I have looked at the folder/model_parameters/trainingmodel_cd_cont_1000_x
I have 6 different folders for x going from 1-8
However, none of them except one contains feat.param file.
Am I suppose to copy to each folder?

As for the frame rate, I am actually doing acoustic model adaptation and see
how different frame rate can lead to better performance for different speed of
speech files.
If this is true, is there any parameter that I can try to change?

Thank you

[Nickolay V. Shmyrev]

Am I suppose to copy to each folder?

Why do you think you need to copy anything? You have one folder with the final
model (without _x) and this folder contains feat.params. And this folder is
used during decoding according to configuration in sphinx_decode.cfg. And you
need to edit the file in this folder.

As for the frame rate, I am actually doing acoustic model adaptation and see
how different frame rate can lead to better performance for different speed of
speech files. If this is true,

Frame rate is not related to any kind of speed of speech. It's only make sense
to change frame rate when you decode lossy speech encoded by VoIP codec. For
example many codecs have frame size 30ms, then your feature extraction should
also preferably have same frame size (frame rate is 33).

is there any parameter that I can try to change?

I don't see why do you need to change any parameter.

[Suong-Sun Hong]

Hi,

Thank you for your last reply.
I think I didn't make it clear.

So when I'm running ./scripts_pl/make_feats.pl -ctl /etc/train.fileids
it looks for the 'feat.params' in etc folder.
Therefore I changed the 'feat.params' in etc folder (I also changed the one in
the model folder you told me),
however, it still goes back to default value every time I run that script
file.
What am I doing wrong here?

Thanks

[Nickolay V. Shmyrev]

it looks for the 'feat.params' in etc folder.

No, it doesn't use this file. Instead, it uses the file from
model_parameters/model_name/feat.params.

What am I doing wrong here?

You changed wrong file