[Subject Prev][Subject Next][Thread Prev][Thread Next][Date Index][Thread Index]

[hts-users:01210] Re: how to set the hlist.conf


Hi,

paminy wrote (2008/03/10 17:54):

1. I gained the data as follow:

------ Source: /hts/HTS-demo_CMU-ARCTIC-SLT2/data/cmp/cmu_us_arctic_slt_a0001.cmp ------
  Sample Bytes:  312      Sample Kind:   USER
  Num Comps:     78       Sample Period: 5000.0 us
  Num Samples:   669      File Format:   HTK
------------------------------------ Samples: 0->-1 ------------------------------------ 0: 2.467 1.391 0.333 0.366 0.175 0.146 0.086 0.140 -0.089 0.144 0.055 0.018 0.016 0.027 -0.028 0.055 -0.006 0.025 0.011 -0.024 -0.044 0.054 -0.024 0.008 -0.037 1.191 0.601 0.212 0.256 0.102 0.127 0.063 0.118 -0.029 0.119 0.024 0.058 -0.003 0.064 0.003 0.033 0.020 0.011 0.032 0.006 -0.024 0.042 0.007 0.010 0.017 -2.552 -1.580 -0.241 -0.220 -0.146 -0.039 -0.047 -0.045 0.119 -0.048 -0.063 0.080 -0.037 0.075 0.061 -0.044 0.051 -0.029 0.042 0.060 0.039 -0.022 0.063 0.003 0.107-10000000000.000-10000000000.000-10000000000.000
......

I am sorry to tell you that I can't understand the data in the cmu_us_arctic_slt_a0001.cmp file.Could you help me and interpret them . Dose it means the first window MGC MERGE with the first window f0 ?

First 25 dimensions are mel-cepstrum, 26--50 dimensions are delta mel-cepstrum,
51--75 dimensions are delta-delta mel-cepstrum, and 76-th, 77-th, and 77-th dimensions are
log F0, delta log F0, and delta-delta log F0, respectively.
-10000000000.000 denotes log(0.0), it means unvoiced.
You may find log F0 and their dynamic features at voiced/unvoiced boundaries has a bit strange like 4.5 -10000000000.000 -10000000000.000.
We cannot calculate dynamic features at voiced/unvoiced boundaries because dynamic features
are calculated from their neighboring frames.

2. I am a novice in Speech Recognition , and the HTS is difficulty for me(you know that there is no HTSbook ...) .How can I improve my understanding with the HTS.

Please read papers about HTS.
A number of papers of HTS have been published.
Some of them are available at the publication page of the HTS website:

http://hts.sp.nitech.ac.jp/?Publications

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
------------------------------------------------

Follow-Ups
[hts-users:01211] Problems in algorithm of parameter generation in HGen.c and HTS_mplg.c, QHE
References
[hts-users:01208] Re: how to set the hlist.conf, Heiga ZEN (Byung Ha CHUN)
[hts-users:01207] how to set the hlist.conf, paminy
[hts-users:01209] Re: how to set the hlist.conf, paminy