소프트웨어 (Software)

[홈][기술][수학이론][소프트웨어][사용법][링크][Q&A]

To download the programs, go to the Download page.

kPhonetica

Speech waveform editing

Spectrogram display

Energy/Pitch/Formant contours

Manual/Automatic labeling

Formant chart, Q-tone level (QTL)

For detailed information, see the Manual page.

kWaves

Speech waveform editing

Spectrogram display

Manual labeling

Automatic aligning

Pitch detection

Speech analysis

Feature extraction

For detailed information, see the Manual page.

ezCSR (Easy Continuous Speech Recognizer)

Functions:

        Can recognize isolated words or continuous speech

        Can support two kinds of grammar: n-gram, finite state network (FSN)

        Can run multiple speech recognizers with different vocabulary and grammar

        Can configure speech recognizers flexibly for most of purposes

        Can handle multiple pronunciation

        Support input speech from sound devices or files

        Support MFCC, FBANK, LPCC, PLP feature

        Support continuous hidden Markov modeling

        Support FSN-based search and tree-based search

        Support Hangul (Korean language) dictionary

        Support decision tree-based  and mapping-based state sharing

        Support (reduced) file formats compatible with the HTK

        Support automatic speech detection

 

Systems/Compilers Supported:

        Windows 2000 Professional/Microsoft Visual C++ 6.0

        FreeBSD/GNU C++ v2.95.2 (except audio I/O module)

 

This software requires the Easy Matrix Template Library (ezMTL).

ezMTL (Wasy Matrix Templay Library)

C++ template library for matrix and vector algebra

 

ISytems/Compilers:

    Windows 32/Visual C++ 6.0

    FreeBSD/GNU C++ 2.95.2

 

Functions:

       - Most of elementary matrix operations (+, -, *, /, transpose, conjugate, ...)

       - Some of basic statistical operations (sum, mean, var, std, skewness, kurtosis, ...)

       - Checking matrix properties (symmetric, Hermitian, positive definite, diagonal, triangular, zero, identity, ...)

       - Cholesky factorization for square real/complex symmetric positive definite matrices

       - LU factorization for square real/complex matrices

       - QR factorization for general real/complex matrices

       - Singular value decompostion for general real matrices

       - Hessenberg form for general real/complex matrices

       - Schur decomposition for general real/complex matrices

       - BKP decomposition for general real matrices

       - Eigenvectors and Eigenvalues for symmetric/Hermitian matrices

       - Eigenvectors and Eigenvalues for general real/complex matrices

       - Inverse, pseudoinverse, determinant, p-norm

       - Rank, orthogonal/null space

       - Condition numbers

       - Solving linear equations

       - Matrix exponential, logarithm, square root, power

       - Random number generation (uniform, normal, exponential, gamma, beta, t, fisher, chi-square, binomial, poisson, ...)

 

Notes:

This library was designed to make MATLAB codes run in C++. Most of method names are compatible to MATLAB commands except the first character is capital letter. This library is partly based on the Meschach Library.

hantagger

Hangeul part-of-speech (POS) tagging tool

This is a quick-and-dirty program to segment hangeul sentences
into morphemes. It uses the definition of morpheme developed at KAIST, Korea.
This program is useful when a simple Korean tagging system is desired.
I cannot gurantee the accuracy or correctness of the program.
I open the program because it will be useful in some applications.

Applications targeted:
   Morpheme-dependent processing of Korean text-to-pronunciation conversion
   Simple tagging for speech synthesis systems
   Quick-and-dirty implementation of Korean parsers

The followings are some tagged examples.
in=강의의 수준 자라고 있다
out=강의/ncn+의/jcm 수준/ncn 자라/pvg+고/ecx 있/px+다/ef
in=이것은 자라고 생각한다 갈 수 없는 나라입니다
out=이것/npd+은/jxt 자라/pvg+고/ecc 생각/ncpa+하/xsv+ㄴ다/ecs 가/pvg+ㄹ/etm 수/nbn 없/paa+는/etm 나라/ncn+이/jp+ㅂ니다/ef
in=강의의 단옷골 갈수각 갈수각은 간다
out=강의/ncn+의/jcm 단옷골/nq 갈수각/nq 갈수각/nq+은/jxt 가/px+ㄴ다/ef
in=권오욱은 고유명사 테스트합니다
out=권오욱/nq+은/jxt 고유/ncn+명사/ncn 테스트/ncpa+하/xsv+ㅂ니다/ef

hanttp

This is a quick-and-dirty program to convert Korean texts into pronunciations.
The program takes into account part-of-speech information of eojeols
and needs a Korean tagging system included in the current distribution.
If you are sensitive to the speed of conversion, you should revise
the program for your own application.
I open the program because it will be useful in some applications.

Applications:
   Pronunciation generation in variable-vocabulary speech recognizers
   Automatic generation of pronunciation dictionary for large vocabulary speech recognizers
   Text preprocessing for speech synthesis
   
The followings are some examples of the conversion (version 1.0).

in : 의학 의사 회의실 가정의학 감다 갈 것 같은 사람의 앉히다 굳히다 같이 강의의 혜례예혜 밝은 ㄹ것같 측심의 가능성 의의를 둔다
out: 의학 의사 훼이실 가정의학 감따 갈 꺼ㄷ 가튼 사라메 안치다 구치다 가치 강이에 헤례예헤 발근 ㄹ꺼ㄷ까ㄷ 측씨미 가능썽 의이를 둔다
in : 신라의의사신라독립만세국회밝은세상희망회사훼방군
out: 실라이이사실라동님만세구퀘발근세상히망훼사훼방군

split2eojeol

This program is to segment hangul sentences into eojeols.
Usage:
split2eojeol.exe [-sentence "A"] [-compare "A" "B" ...] [-compound "A"]
[-print_org] [-minWordLen 1] [-maxWordLen 12] [-lm LM_file] [-vocab vocab_file]
[-beamSize 10] [-penInsertion 0.0] [-splitTh 1] [-dur_wt 1.0] [-keepSpace] < file
To see detailed message, use the option -h.
Options:
       -h|help          help
       -sentence "A"    split the sentence A
       -compare "A" "B" compare between A and B
       -compound "A"    split a compound noun A
       -print_org       print original input
Advanced options:
       -minWordLen      minimum word length after spacing
       -maxWordLen      maximum word length after spacing
       -lm              ARPA language model file
       -vocab           non-compound noun file
       -beamSize        beam size
       -penInsertion    penalty for space insertion
       -splitTh         split input longer than or equal to the threshold
       -dur_wt          weight for word length score
       -keepSpace       keep original spaces
Examples:
       split2eojeol < Column.txt
       split2eojeol -sentence "이것을띄어쓰기하면어떻게될까요"
       split2eojeol -compare "어떤게맞을까요" "어떤게 맞을까요" "어떤 게 맞을까요"
       split2eojeol -compound "기초지방자치단체행정전산화"

TIMIT48

The TIMIT is a basic speech database in speech recognition fields. It is often used not ony for speaker independent phoneme recognition but also for speaker recognition, microphone adaptation and channel adaptation.
When one wants to experiment using the HTK with the TIMIT database, there are many steps and parameters to be set properly.
The attached script can be used to do all the required steps and parameter setting in context-independent and context-dependent phoneme recognition with just a single command.

To download the above programs, go to the Download page.

[홈][기술][수학이론][소프트웨어][사용법][링크][Q&A]

© 2009 Oh-Wook Kwon.