[hts-users:03352] Re: Crash when using HTS-2.1 training HTS-demo_CMU-ARCTIC-SLT

[Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>]

Hi,

12GB is enough to run the demo script.
Do you use special CPU?
Could you try to use HTS-2.2 and its demo scripts?

Regards,
Keiichiro Oura


2012/6/15 ChunRong Li <qcclcr@xxxxxxxxx>:
> Hi,   Keiichiro Oura
>     The memory size is 12GB in my server computer (Ubuntu).
>
> $ cat  /proc/meminfo
> MemTotal:       12333272 kB
> MemFree:        11755332 kB
> Buffers:           66388 kB
> Cached:           124544 kB
> SwapCached:            0 kB
> .....
>
> 2012/6/15 Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>
>>
>> Hi,
>>
>> A large amount of memory is required for BuildTree step.
>> Please, let me know the memory size of your computer.
>>
>> Regards,
>> Keiichiro Oura
>>
>>
>> 2012/6/15 ChunRong Li <qcclcr@xxxxxxxxx>:
>> > Hi,everyone
>> >    I  download HTK-3.4.tar.gz and HTS-2.1_for_HTK-3.4.tar.bz2,  I
>> > patched
>> > HTS into HTK,  and install HTK successfully。
>> > And I used HTS-demo_CMU-ARCTIC-SLT.tar.bz2 for test,   during training ,
>> > the
>> > computer rebooted,  nothing prompted.
>> > And after  I debuged with gdb  I find that  computer reboot in doing
>> > function BuildTree( ).   But I still not found the reason.
>> >
>> >
>> > Following is the information in log:
>> >
>> > HHEd
>> >    38765/38765 Models Loaded [7 states max, 2 mixes max]
>> >
>> > //  load stats files
>> > RO 0.00 ' '
>> >  Setting outlier threshold for clustering
>> >  RO->LS
>> > /home/hmm/work/HTS-demo_CMU-ARCTIC-SLT/stats/qst001/ver1/cmp.stats
>> >    and loading state occupation stats
>> >    Stats loaded for 38765 models
>> >    Mean Occupation Count = 3.491811
>> >
>> > TR 0
>> >   Adjusting trace level
>> >
>> > TR 3
>> >   Adjusting trace level
>> >
>> > //  construct decision tree
>> > TB 0.00 mgc_s2_ {*.stats[2].streams[1-1]}
>> >   Tree based clustering based on MDL criterion, threshold=8.594570e+02
>> >  TB: Stats 38765->232 [0.6%]  { 77530->516 [0.7%] total }
>> >
>> > TB 0.00 mgc_s4_  {*.state[4].stream[1-1]}
>> >  Tree based clustering based on MDL criterion, threshold=8.954802e+02
>> >
>> > Chunrong LI
>> >
>>
>