Hi there,
I know KALDI is able to use user defined features so I tried to use my own frontend and generated my own features but when I used $delta_opts in feats="ark,s,cs:apply-cmvn..." on monophone to eliminate the effect of deltas, the accuracy is dramatically lower in compare to HTK. IS there any reason for that? Is it the only thing that I need to do?
Next on train_deltas.sh for eliminating the deltas, (this time for triphone) I added "--delta-opts "--delta-order=0"" to feats="ark,s,cs:apply-cmvn..." and it made the job failed with this error log:
Add deltas (typically to raw mfcc or plp features
Usage: add-deltas [options] in-rspecifier out-wspecifier
Options:
--delta-order : Order of delta computation (int, default = 2)
--delta-window : Parameter controlling window for delta computation (actual window size for each delta order is 1 + 2*delta-window-size) (int, default = 2)
--truncate : If nonzero, first truncate features to this dimension. (int, default = 0)
Standard options:
--config : Configuration file to read (this option may be repeated) (string, default = "")
--help : Print out usage message (bool, default = false)
--print-args : Print the command line arguments (to stderr) (bool, default = true)
--verbose : Verbose level (higher->more logging) (int, default = 0)
LOG (acc-tree-stats:main():acc-tree-stats.cc:149) Accumulated stats for 0 files, 0 failed due to no alignment, 0 failed for other reasons.
LOG (acc-tree-stats:main():acc-tree-stats.cc:152) Number of separate stats (context-dependent states) is 0
WARNING (acc-tree-stats:Close():kaldi-io.cc:446) Pipe gunzip -c exp/mono0a_ali/ali.1.gz| had nonzero return status 13
WARNING (acc-tree-stats:Close():kaldi-io.cc:446) Pipe apply-cmvn --utt2spk=ark:data/train_si84_half/split8/1/utt2spk scp:data/train_si84_half/split8/1/cmvn.scp scp:data/train_si84_half/split8/1/feats.scp ark:- | add-deltas --delta-opts --delta-order=0 ark:- ark:- | had nonzero return status 65280
Accounting: time=0 threads=1
Ended (code 1) at Tue Jul 7 16:11:03 EDT 2015, elapsed time 0 seconds
Could anybody help me in this regard?
Thanks in advanced
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
The form "--delta-opts "--delta-order=0"" would be used on the command
line of train_deltas.sh, not added as a modification to the script.
You'd just want --delta-order=0 if modifying the script. The option
format of the scripts requires a space, but of the c++ programs
requires an "=" sign.
Dan
Hi there,
I know KALDI is able to use user defined features so I tried to use my own
frontend and generated my own features but when I used $delta_opts in
feats="ark,s,cs:apply-cmvn..." on monophone to eliminate the effect of
deltas, the accuracy is dramatically lower in compare to HTK. IS there any
reason for that? Is it the only thing that I need to do?
Next on train_deltas.sh for eliminating the deltas, (this time for triphone)
I added "--delta-opts "--delta-order=0"" to feats="ark,s,cs:apply-cmvn..."
and it made the job failed with this error log:
Add deltas (typically to raw mfcc or plp features
Usage: add-deltas [options] in-rspecifier out-wspecifier
Options:
--delta-order : Order of delta computation (int, default = 2)
--delta-window : Parameter controlling window for delta computation (actual
window size for each delta order is 1 + 2*delta-window-size) (int, default =
2)
--truncate : If nonzero, first truncate features to this dimension. (int,
default = 0)
Standard options:
--config : Configuration file to read (this option may be repeated) (string,
default = "")
--help : Print out usage message (bool, default = false)
--print-args : Print the command line arguments (to stderr) (bool, default =
true)
--verbose : Verbose level (higher->more logging) (int, default = 0)
LOG (acc-tree-stats:main():acc-tree-stats.cc:149) Accumulated stats for 0
files, 0 failed due to no alignment, 0 failed for other reasons.
LOG (acc-tree-stats:main():acc-tree-stats.cc:152) Number of separate stats
(context-dependent states) is 0
WARNING (acc-tree-stats:Close():kaldi-io.cc:446) Pipe gunzip -c
exp/mono0a_ali/ali.1.gz| had nonzero return status 13
WARNING (acc-tree-stats:Close():kaldi-io.cc:446) Pipe apply-cmvn
--utt2spk=ark:data/train_si84_half/split8/1/utt2spk
scp:data/train_si84_half/split8/1/cmvn.scp
scp:data/train_si84_half/split8/1/feats.scp ark:- | add-deltas --delta-opts
--delta-order=0 ark:- ark:- | had nonzero return status 65280
Accounting: time=0 threads=1
Ended (code 1) at Tue Jul 7 16:11:03 EDT 2015, elapsed time 0 seconds
Could anybody help me in this regard?
Thanks in advanced
Hi there,
I know KALDI is able to use user defined features so I tried to use my own
frontend and generated my own features but when I used $delta_opts in
feats="ark,s,cs:apply-cmvn..." on monophone to eliminate the effect of
deltas, the accuracy is dramatically lower in compare to HTK. IS there any
reason for that? Is it the only thing that I need to do?
Next on train_deltas.sh for eliminating the deltas, (this time for
triphone) I added "--delta-opts "--delta-order=0"" to
feats="ark,s,cs:apply-cmvn..." and it made the job failed with this error
log:
Add deltas (typically to raw mfcc or plp features
Usage: add-deltas [options] in-rspecifier out-wspecifier
Options:
--delta-order : Order of delta computation (int, default =
2)
--delta-window : Parameter controlling window for delta
computation (actual window size for each delta order is 1 +
2*delta-window-size) (int, default = 2)
--truncate : If nonzero, first truncate features to
this dimension. (int, default = 0)
Standard options:
--config : Configuration file to read (this option
may be repeated) (string, default = "")
--help : Print out usage message (bool, default =
false)
--print-args : Print the command line arguments (to
stderr) (bool, default = true)
--verbose : Verbose level (higher->more logging) (int,
default = 0)
LOG (acc-tree-stats:main():acc-tree-stats.cc:149) Accumulated stats for 0
files, 0 failed due to no alignment, 0 failed for other reasons.
LOG (acc-tree-stats:main():acc-tree-stats.cc:152) Number of separate stats
(context-dependent states) is 0
WARNING (acc-tree-stats:Close():kaldi-io.cc:446) Pipe gunzip -c
exp/mono0a_ali/ali.1.gz| had nonzero return status 13
WARNING (acc-tree-stats:Close():kaldi-io.cc:446) Pipe apply-cmvn
--utt2spk=ark:data/train_si84_half/split8/1/utt2spk
scp:data/train_si84_half/split8/1/cmvn.scp
scp:data/train_si84_half/split8/1/feats.scp ark:- | add-deltas --delta-opts
--delta-order=0 ark:- ark:- | had nonzero return status 65280
Accounting: time=0 threads=1
Ended (code 1) at Tue Jul 7 16:11:03 EDT 2015, elapsed time 0 seconds
Could anybody help me in this regard?
Thanks in advanced
Thanks it worked perfect.
Now on aligning of tri1, it fails with this error message;
run.pl: 8 / 8 failed, log is in exp/tri1_ali_si84/log/align.*.log
but "exp/tri1_ali_si84" directory is not generated. Could you please tell me why it is not generated? or how can I have access to that log?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks it worked perfect.
Now on aligning of tri1, it fails with this error message;
run.pl: 8 / 8 failed, log is in exp/tri1_ali_si84/log/align.*.log
but "exp/tri1_ali_si84" directory is not generated. Could you please tell me
why it is not generated? or how can I have access to that log?
Thanks Dan and Jan for all your helps. The problem was because of a permission issue.
Now for decoding the dimension of my data is tripled.
Here is the error:
ERROR (gmm-latgen-faster:LogLikelihoodZeroBased():decodable-am-diag-gmm.cc:50) Dim mismatch: data dim = 117 vs. model dim = 39
WARNING (gmm-latgen-faster:~HashList():util/hash-list-inl.h:116) Possible memory leak: 1022 != 1024: you might have forgotten to call Delete on some Elems
Do you have any idea what can do this to my data?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
It looks like the delta-opts are not being passed in to the decoding
script- most likely because you hacked the training script instead of
adding "--delta-opts "--delta-order=0"" on the command line of the
train_deltas.sh script.
Dan
Thanks Dan and Jan for all your helps. The problem was because of a
permission issue.
Now for decoding the dimension of my data is tripled.
Here is the error:
ERROR
(gmm-latgen-faster:LogLikelihoodZeroBased():decodable-am-diag-gmm.cc:50) Dim
mismatch: data dim = 117 vs. model dim = 39
WARNING (gmm-latgen-faster:~HashList():util/hash-list-inl.h:116) Possible
memory leak: 1022 != 1024: you might have forgotten to call Delete on some
Elems
Do you have any idea what can do this to my data?
On monophone script, this option doesn't exist by default so I just modified feats on train_mono.sh that is:
feats="ark,s,cs:apply-cmvn $cmvn_opts --utt2spk=ark:$sdata/JOB/utt2spk scp:$sdata/JOB/cmvn.scp scp:$sdata/JOB/feats.scp ark:- | add-deltas $delta_opts ark:- ark:- |"
and defined delta_opts on the beginning of code then called this script from command line by "--delta-opts "--delta-order=0"". Is this wrong?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
If you put the string
--delta-order=0
into a file exp/mono/delta_opts (or whatever the directory is) the
decoding should work.
You need to be willing to look a little into the scripts. this is not
rocket science.
Dan
On monophone script, this option doesn't exist by default so I just modified
feats on train_mono.sh that is:
feats="ark,s,cs:apply-cmvn $cmvn_opts --utt2spk=ark:$sdata/JOB/utt2spk
scp:$sdata/JOB/cmvn.scp scp:$sdata/JOB/feats.scp ark:- | add-deltas
$delta_opts ark:- ark:- |"
and defined delta_opts on the beginning of code then called this script from
command line by "--delta-opts "--delta-order=0"". Is this wrong?
Hi there,
I know KALDI is able to use user defined features so I tried to use my own frontend and generated my own features but when I used $delta_opts in feats="ark,s,cs:apply-cmvn..." on monophone to eliminate the effect of deltas, the accuracy is dramatically lower in compare to HTK. IS there any reason for that? Is it the only thing that I need to do?
Next on train_deltas.sh for eliminating the deltas, (this time for triphone) I added "--delta-opts "--delta-order=0"" to feats="ark,s,cs:apply-cmvn..." and it made the job failed with this error log:
acc-tree-stats --ci-phones=1:2:3:4:5:6:7:8:9:10:11:12:13:14:15 exp/mono0a_ali/final.mdl "ark,s,cs:apply-cmvn --utt2spk=ark:data/train_si84_half/split8/1/utt2spk scp:data/train_si84_half/split8/1/cmvn.scp scp:data/train_si84_half/split8/1/feats.scp ark:- | add-deltas --delta-opts --delta-order=0 ark:- ark:- |" "ark:gunzip -c exp/mono0a_ali/ali.1.gz|" exp/tri1/1.treeacc
Started at Tue Jul 7 16:11:03 EDT 2015
acc-tree-stats --ci-phones=1:2:3:4:5:6:7:8:9:10:11:12:13:14:15 exp/mono0a_ali/final.mdl 'ark,s,cs:apply-cmvn --utt2spk=ark:data/train_si84_half/split8/1/utt2spk scp:data/train_si84_half/split8/1/cmvn.scp scp:data/train_si84_half/split8/1/feats.scp ark:- | add-deltas --delta-opts --delta-order=0 ark:- ark:- |' 'ark:gunzip -c exp/mono0a_ali/ali.1.gz|' exp/tri1/1.treeacc
apply-cmvn --utt2spk=ark:data/train_si84_half/split8/1/utt2spk scp:data/train_si84_half/split8/1/cmvn.scp scp:data/train_si84_half/split8/1/feats.scp ark:-
Add deltas (typically to raw mfcc or plp features
Usage: add-deltas [options] in-rspecifier out-wspecifier
Options:
--delta-order : Order of delta computation (int, default = 2)
--delta-window : Parameter controlling window for delta computation (actual window size for each delta order is 1 + 2*delta-window-size) (int, default = 2)
--truncate : If nonzero, first truncate features to this dimension. (int, default = 0)
Standard options:
--config : Configuration file to read (this option may be repeated) (string, default = "")
--help : Print out usage message (bool, default = false)
--print-args : Print the command line arguments (to stderr) (bool, default = true)
--verbose : Verbose level (higher->more logging) (int, default = 0)
Command line was: add-deltas --delta-opts --delta-order=0 ark:- ark:-
ERROR (add-deltas:Read():parse-options.cc:375) Invalid option --delta-opts
ERROR (add-deltas:Read():parse-options.cc:375) Invalid option --delta-opts
[stack trace: ]
kaldi::KaldiGetStackTrace()
kaldi::KaldiErrorMessage::~KaldiErrorMessage()
kaldi::ParseOptions::Read(int, char const const)
add-deltas(main+0x1c1) [0x451f15]
/lib64/libc.so.6(__libc_start_main+0xfd) [0x31b6e1ed5d]
add-deltas() [0x451c99]
LOG (acc-tree-stats:main():acc-tree-stats.cc:149) Accumulated stats for 0 files, 0 failed due to no alignment, 0 failed for other reasons.
LOG (acc-tree-stats:main():acc-tree-stats.cc:152) Number of separate stats (context-dependent states) is 0
WARNING (acc-tree-stats:Close():kaldi-io.cc:446) Pipe gunzip -c exp/mono0a_ali/ali.1.gz| had nonzero return status 13
WARNING (acc-tree-stats:Close():kaldi-io.cc:446) Pipe apply-cmvn --utt2spk=ark:data/train_si84_half/split8/1/utt2spk scp:data/train_si84_half/split8/1/cmvn.scp scp:data/train_si84_half/split8/1/feats.scp ark:- | add-deltas --delta-opts --delta-order=0 ark:- ark:- | had nonzero return status 65280
Accounting: time=0 threads=1
Ended (code 1) at Tue Jul 7 16:11:03 EDT 2015, elapsed time 0 seconds
Could anybody help me in this regard?
Thanks in advanced
The form "--delta-opts "--delta-order=0"" would be used on the command
line of train_deltas.sh, not added as a modification to the script.
You'd just want --delta-order=0 if modifying the script. The option
format of the scripts requires a space, but of the c++ programs
requires an "=" sign.
Dan
On Tue, Jul 7, 2015 at 1:37 PM, Eric Camalo roozbeh412@users.sf.net wrote:
Re the second question:
the command line should not be add-deltas --delta-opts --delta-order=0
it should be just add-deltas--delta-order=0
y
On Tue, Jul 7, 2015 at 4:37 PM, Eric Camalo roozbeh412@users.sf.net wrote:
Thanks it worked perfect.
Now on aligning of tri1, it fails with this error message;
run.pl: 8 / 8 failed, log is in exp/tri1_ali_si84/log/align.*.log
but "exp/tri1_ali_si84" directory is not generated. Could you please tell me why it is not generated? or how can I have access to that log?
Hm.
I suspect that that directory does exist but you did not see it
somehow. Or possibly a weird permissions or umask issue.
Dan
On Tue, Jul 7, 2015 at 8:41 PM, Eric Camalo roozbeh412@users.sf.net wrote:
Thanks Dan and Jan for all your helps. The problem was because of a permission issue.
Now for decoding the dimension of my data is tripled.
Here is the error:
ERROR (gmm-latgen-faster:LogLikelihoodZeroBased():decodable-am-diag-gmm.cc:50) Dim mismatch: data dim = 117 vs. model dim = 39
WARNING (gmm-latgen-faster:~HashList():util/hash-list-inl.h:116) Possible memory leak: 1022 != 1024: you might have forgotten to call Delete on some Elems
Do you have any idea what can do this to my data?
It looks like the delta-opts are not being passed in to the decoding
script- most likely because you hacked the training script instead of
adding "--delta-opts "--delta-order=0"" on the command line of the
train_deltas.sh script.
Dan
On Wed, Jul 8, 2015 at 10:22 AM, Eric Camalo roozbeh412@users.sf.net wrote:
On monophone script, this option doesn't exist by default so I just modified feats on train_mono.sh that is:
feats="ark,s,cs:apply-cmvn $cmvn_opts --utt2spk=ark:$sdata/JOB/utt2spk scp:$sdata/JOB/cmvn.scp scp:$sdata/JOB/feats.scp ark:- | add-deltas $delta_opts ark:- ark:- |"
and defined delta_opts on the beginning of code then called this script from command line by "--delta-opts "--delta-order=0"". Is this wrong?
If you put the string
--delta-order=0
into a file exp/mono/delta_opts (or whatever the directory is) the
decoding should work.
You need to be willing to look a little into the scripts. this is not
rocket science.
Dan
On Wed, Jul 8, 2015 at 11:07 AM, Eric Camalo roozbeh412@users.sf.net wrote: