I checked the list folder in the HTS engine under full_all and i found the phoneme "aa" model with 3 context
what are the minimal numbers of training data required for the phoneme "aa" to pass the HInit Function correclty (error number is 2021)
thanks
From: "Heiga ZEN (Byung Ha CHUN)" <zen@xxxxxxxxxxxxxxx>
Reply-To: hts-users@xxxxxxxxxxxxxxx
To: hts-users@xxxxxxxxxxxxxxx
Subject: [hts-users:00594] Re: HInit error
Date: Thu, 29 Mar 2007 00:19:41 +0900
>Hi,
>
>Tamer Fares wrote:
>
>>in order to generates a new arabic voice
>>
>>i did the following
>>
>>1.i recorded 50 sentences.raw
>>
>>2. I made the label mono file for each sentence
>>
>>3.i prepared the full context label files for each sentence in the
>>format p1^p2+p3-p4=p5 only
>>
>>i put them in the label folder in a new HTS demo application
>>
>>i compiled it then it extracts the mcp and f0 then it calculates
>>the cmp files
>>
>>but when it starts to run the HInit, it gave me an error (Too few
>>parameters observation)
>>
>>this error due to the small numbers of sentences or due to that i
>>taked the format of the full context like above
>
>This error was caused by the small numbers of training sentences.
>Please check whether each model has sufficient training samples or
>not.
>
>>i didn't make the questionaires file . from where i can build my
>>own one for my language.there is any documents where i can know how
>>to design it
>
>There is no document but this topic has been discussed many times in
>this mailing list.
>So you can find related mails using the hts-users ML archive.
>
>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
>------------------------------------------------
>