[hts-users:01009] 答复: [hts-users:01008] Re: about computation load

["lixiulin" <lixiulin@xxxxxxxxxxxxxxxxxxxxx>]

Dear Heiga,

Thanks for your reply.

"12xPentiumD 3.2GHz machines (4GB RAM) " means a cluster with 12 machines, and each machine has 4G ram, yes?

What do tricks mean? "-r" option is one of them? if "-r" is used, how about the memory consuming?


Best regards,

Li Xiulin


-----原始邮件-----
发件人: Heiga ZEN (Byung Ha CHUN) [mailto:zen@xxxxxxxxxxxxxxx]
发送时间: 2007年11月26日 13:13
收件人: hts-users@xxxxxxxxxxxxxxx
主题: [hts-users:01008] Re: about computation load


Hi,

lixiulin wrote (2007/11/26 13:59):

>>> I am planing to train a large speech corpus with 400,000 syllables or so. 
>>> I knew from mailing list that a 64-bit system is helpful, but how about the computer/machine?  
>>> A server or a PC? How about the time for training such a DB?
>> 
>> Can you tell me how large your database is in hours?
>
> The speech data is about 20 hours or so.

I used 12xPentiumD 3.2GHz machines (4GB RAM) to train 5 hours of speech data, it took a day. 
So if you have similar computational resources, it may take 4 days.

Note that training 5 hours of speech data consumed about 3GB RAM, 
so training 20 hours of speech data may require 15GB (if you don't use any trick to reduce memory).

Regards,

Heiga ZEN (Byung Ha CHUN)

-- 
------------------------------------------------
 Heiga ZEN     (in Japanese pronunciation)
 Byung Ha CHUN (in Korean pronunciation)

 Department of Computer Science and Engineering
 Nagoya Institute of Technology
 Gokiso-cho, Showa-ku, Nagoya 466-8555 Japan

 http://www.sp.nitech.ac.jp/~zen
------------------------------------------------