|
hi，
Maybe the examples in your dataset are too
short to be picked up. what is your state number? You have to make sure that the
durations marked in your label file are long enough, so that the segments have
enough frames to allocate to each state. The number of
frames in the segment must be bigger than the state
number.
There are more than 115 examples of
phoneme, but maybe all of
them are too short.
For example, if you are using 5-state
hmms, and the frame shift(not frame length) is 80, then in your label file,
at least one example whoes duration is long than 5*80*16000 should be
found. Please check it.
孙敬伟
2008-04-18 发件人： Zuko
Fani
发送时间：
2008-04-18 16:06:39
收件人：
hts-users@xxxxxxxxxxxxxxx
抄送：
主题： [hts-users:01341] Re:
ERROR [+2121] HInit: Too FewObservation Sequences [0]
Hi
There are more than 115 examples of the phoneme in the dataset.
regards
Zuko
> > > Simon King <Simon.King@xxxxxxxx
> 04/17/08 4:39 PM > > >
Zuko Fani wrote:
> Hi
>
> HTS can't pick up segments because my labs are less than 5 frames where 1 frame length is 25ms. I tried reducing FRAMELEN = 160 # Frame length in point (400 = 16000 * 0.01) from Makefile script. But still I'm receiving the same error ERROR [+2121] HInit: Too Few Observation Sequences [0]. Is there another parameter I can change to accommodate my labels for phoneme boundaries.
That error message might actually mean that you have no examples of a
particular segment type in your database.
Simon
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
--
This message is subject to the CSIR's copyright terms and conditions, e-mail legal notice, and implemented Open Document Format (ODF) standard.
The full disclaimer details can be found at http://www.csir.co.za/disclaimer.html.
This message has been scanned for viruses and dangerous content by MailScanner,
and is believed to be clean. MailScanner thanks Transtec Computers for their support. |