Explanation of Parameters to sphinx3 livepretend

From Sphinx

  • Acoustic model: Get phonemes from raw audio 
-hmm $SPHINX_ROOT/sphinx3/model/hmm/hub4_cd_continuous_8gau_1s_c_d_dd

file contains statistical characteristics (HMMs - hidden Markov models) for each phoneme: ay, sh, ah,...

use SphinxTrain to generate it from {RawAudio + transcript}. #here's how


  • Lexical model: Phonetic dict - Get words from phonemes 
-dict $SPHINX_ROOT/lm_giga_5k_nvp_3gram/lm_giga_5k_nvp.sphinx.dic
-fdict $SPHINX_ROOT/lm_giga_5k_nvp_3gram/lm_giga_5k_nvp.sphinx.filler

Typical line of blah.dic: aviation ey v iy ey sh ah n Filler is for sounds like coughing, umming etc


  • Language Model: %prob of each word & pairs/triples 
-lm $SPHINX_ROOT/lm_giga_5k_nvp_3gram/lm_giga_5k_nvp_3gram.arpa.DMP

Comes from blah.arpa textfile - looks like this.

Have to convert .arpa to .DMP so Sphinx can use it. 

$ lm3g2dmp $SPHINX_ROOT/lm_giga_5k_nvp_3gram/lm_giga_5k_nvp_3gram.arpa $SPHINX_ROOT/lm_giga_5k_nvp_3gram


More details at http://cmusphinx.sourceforge.net/sphinx3/doc/s3_description.html#sec_decoverview


[edit]

here's how

Can someone help on this section? I'm trying to pull it out of http://www.speech.cs.cmu.edu/sphinx/tutorial.html

under dev.........

Have to feed these into SphinxTrain:

  1. The acoustic signals for everything you want to train - .WAV files
  2. The corresponding transcript file
  3. dictionary (language & filler)

and SphinxTrain will chunk out a Model-Index file.

(RawAudio Tarball of CMU students speaking random words & numbers into mic)

  • Convert into set of 13D feature-vectors (MFCCs) 
perl scripts_pl/make_feats.pl  -ctl etc/rm1_train.fileids
  • Now process them some more: 
perl scripts_pl/RunAll.pl
  • This populates 
./model_parameters/rm1.cd_cont_1000_8/(your HMMs)

(ie the parameters of the final 8 Gaussian/state 3-state CD-tied acoustic models (HMMs) with 1000 tied states ) 

./model_architecture/rm1.1000.mdef

Model-Index file for these models (used by the system to associate the appropriate set of HMM parameters with the HMM for each sound unit you are modeling.)

DECODING

Retrieved from "http://sphinx.subwiki.com/sphinx/index.php/Explanation_of_Parameters_to_sphinx3_livepretend"
related