Alignment script issues

[Max C]

Hiya,

I’m building a DNN-system for a possible undergraduate dissertation and I’ve just started so I’m a bit behind some of the other students. When I run the ‘do_alignment_dnn’ script I am shown the following error(s):

CHANGING DIRECTORY to .
CHECKING FOR FILES
CREATING DICT
CREATING MASTER MODEL FILE (MMF)
CALCULATING VARIANCE FLOOR using HCompV
GENERATING MODELS
../scripts/do_alignment_dnn: line 48: hmm0/$m: ambiguous redirect
ERROR [+5010] InitSource: Cannot open source file hmm0/?
ERROR [+7010] LoadHMMSet: Can’t find file
ERROR [+2628] Initialise: LoadHMMSet failed
FATAL ERROR – Terminating program HHEd
EMBEDDED REESTIMATION OF MODEL PARAMETERS (ITERATION 0 of 5)
ERROR [+5010] InitSource: Cannot open source file hmm0/MMF
ERROR [+7010] LoadAllMacros: Can’t open file
ERROR [+7032] CheckOptions: options not set in HMM Set
HMM Def Error: LoadAllMacros: CheckOptions failed at line 1/col 0/char -1 in hmm0/vFloors
ERROR [+7050] HMError:
ERROR [+7050] LoadHMMSet: Macro name expected
ERROR [+2321] Initialise: LoadHMMSet failed
FATAL ERROR – Terminating program HERest
Re-estimation 0 failed, Aborted!

Looking through the directories after the error I can see that the MMF has not been created, and I think this might be the main issue. However I’ve struggled with this for a bit and cannot figure out why I am getting this error.

[Felipe E]

Hi Max,

Check in the file “do_alignment_dnn” that your phone_list and train.scp are correct. Also, check that the output of the line 42 (HCompV -C resources/CONFIG_for_training -f 0.01 -m -S train.scp -M hmm0 proto/7states) is correct.

Thanks,

Felipe

[Max C]

Hi Felipe,

I’ve had a look at my train.scp and phone_list files and this is what they have come out as:

train.scp

<pre class=”lang:bash”>
/Users/s1348525/Documents/ss_dnn/alignment/mfcc/arctic_a0001.mfcc
/Users/s1348525/Documents/ss_dnn/alignment/mfcc/arctic_a0002.mfcc
/Users/s1348525/Documents/ss_dnn/alignment/mfcc/arctic_a0003.mfcc
…

phone_list

<pre class=”lang:bash”>
#
?
@
@@r
…
z
zh

As far as I can tell these are working as they should, there is nothing that seems out of the ordinary with them.

Regarding the output of line 42, the contents of photo/7states seem a bit hard to read, an extract of it is:

proto/7states

<pre class=”lang:bash”>
<State> 2
<Mean> 39
0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
<Variance> 39
1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0
1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0
…

The only other possible issue off the top of my head is that phone_list and train.scp are in the incorrect directory. Both are currently sitting in ss_dnn/alignment. Is this correct?

[Felipe E]

Hi Max,

I have some ideas/questions:

– What is the value of $m when it fails?
– What is inside the directory /hmm0?
– Check the permissions of the directory.

Thanks,

Felipe

[Max C]

Hi,

So this response is a bit delayed but haven’t had a chance to get back into the lab until now. I noticed that the error I was getting referred to a file called /hmm0/? – in the hmm0 directory there is a file for each of the phones in the dictionary (unilex-edi in this case), however for some reason there was no file for the glottal stop “?”. So the do_alignment_dnn script is trying to open this file which is not there, however there is a file for the rest of the phones.

The utts_to_mlfs.sh file creates the phone_list so I don’t think the issue is here. Rather I think the problem is occurring in the do_alignment_dnn script, for some reason the script isn’t creating an /hmm0/? file, but does so for the other phones in the dictionary.

In order to test if the issue solely occurred with the “?” phone, I created a unit-selection voice using the unilex-rpx dictionary instead of unilex-edi. Once that voice was created, I followed all the steps in building the DNN voice, but used the Festival utterance structures from the unilex-rpx voice – which does not contain glottal stops “?”.

Turns out the do_alignment_dnn script works perfectly now. I have a feeling there is an issue with the script creating a file with the character “?” or something along those lines – not exactly sure.

[Etienne D]

This is a bit late, but I encountered the same issue and solved it by creating a “?” file myself (copy-paste-rename). I think it might be due to the fact that exclamation marks are escaped in filenames : “\?”. Another reason not to use weird characters/spaces in filenemes.
I tried to replicate the error independently but it seemed to work fine when not using HTK, so I’d guess the issue is there, but not completely sure.

[Max C]

Hi Etienne,

Thanks for your response. I’ll try this out next time I’m in the lab (next week unfortunately). I had an idea that it was something to do with the character. Script works perfectly fine for every other character representing a phone.

[Author]

Posts