공부삼아 Kaldi copy-feats 프로그램을 뜯어보자

아래 검색어로 찾아보았다.

• how to use kaldi library
• how to use kaldi library on source code
• kaldi library 日本語
• kaldi 日本語 

쓸만한 사이트가 없다. 

직접 소스코드 읽어서 공부해야 함. 

copy-feats 소스코드를 공부하고, 소스코드를 약간 고쳐서 빌드를 시도해보자. 

소스코드는 아래에 

https://github.com/kaldi-asr/kaldi/blob/master/src/featbin/copy-feats.cc 

프린트해서 종이에 메모하였다.

 

gcc가 copy-feats를 빌드하는 과정이 어디 있는지 찾아보자.

src/featbin/Makefile

copy-feat을 빌드하는 Makefile이다.

​Makefile 내용을 캡쳐해서 여기서 주석 달기

SUBDIRS를 빌드하는 부분이다. .PHONY 타겟에 대해 아래를 참고했다.

http://pinocc.tistory.com/131



전부 빌드되는 거라면 나만의 디렉토리를 따로 만들어서 거기다가 copy-feats.cc를 이동하고, src/hybin 을 만들었다.

 make에 수동으로 Makefile 말고 다른 파일을 빌드하도록 전달할 수 있는지 찾아보았다.
(디폴트가 Makefile일 수도 있기 때문)

make 명령 시 target의 이름을 직접 넣어주면 그 타겟의 위치부터 make를 시작함

그러므로 기존 src/Makefile의 내용에 따로 타겟을 만들면 됨.



src/hybin/Makefile을 만들었다.​



copy-feats를 빌드하고, 작동을 확인한다.



아래 오타를 제보하자. 

http://kaldi-asr.org/doc/group__table__group.html#ga0b3ec62400216b77be81739825c9ce28 

아래 인수 중 wxfilename이 rxfilename으로 바뀌어야 함​

Git 사용법 노트

Some features of arrayfire

http://www.evernote.com/l/AdxpZ1EBWE9Ey7Nb2Hy6G6P6hbycLDdQOKY/

JIT compile

lazy evaluation

some tips

Complex double binary to matlab converter

Complex double binary to matlab converter === [Github repository](https://github.com/gogyzzz/complex_double_binary_to_matlab) You can easily load MATLAB complex data in C++(and ArrayFire), and C++ complex data on MATLAB. The structure of binary file is 'interleave Real and Imaginary components' part, demonstrated by [MATLAB R/W COMPLEX NUMBER GUIDE](https://www.mathworks.com/examples/matlab/mw/matlab-ex14013084-write-and-read-complex-numbers) Binary data is stored as 'complex double' type. ### input data type Thus you should put data of complex double type as input(or convert to complex double) But when you use ArrayFire matrix, you can put any type of matrix. ### output data type - complex double in C++ function - c64 in Arrayfire ## References - [MATLAB R/W COMPLEX NUMBER GUIDE](https://www.mathworks.com/examples/matlab/mw/matlab-ex14013084-write-and-read-complex-numbers) - [Arrayfire - Getting Started](http://arrayfire.org/docs/gettingstarted.htm) ## example in C++ ```cpp std::vector<std::complex<double>> a = read_complex64_binary("testmat.complex.interleaved.bin", 2, 2); int nsamples = a.size(); std::cout << "check output\n"; std::copy(a.begin(), a.end(), std::ostream_iterator<std::complex<double>>(std::cout, "\n")); std::cout << "check output from pointer\n"; std::complex<double>* ptr = new std::complex<double>[nsamples]; memcpy(ptr, a.data(), nsamples * sizeof(std::complex<double>)); std::copy(&ptr[0], &ptr[nsamples], std::ostream_iterator<std::complex<double>>(std::cout, " ")); std::cout << std::endl << std::endl; // The filename will be testmat.complex.interleaved.bin write_complex64_binary(ptr, "testmat", 2, 2); // When load binary file, you should put the full name of file. a = read_complex64_binary("A.complex.interleaved.bin", 3, 3, 3); std::copy(a.begin(), a.end(), std::ostream_iterator<std::complex<double>>(std::cout, "\n")); delete[] ptr; ``` ## example in ArrayFire ```cpp array afmat = randn(3, 3, 3, f32); af_print(afmat); std::string realfilename = write_complex64_binary(afmat, "testmat"); std::cout << "the real file name is " << realfilename << std::endl; af_print(read_complex64_binary("A.complex.interleaved.bin", dim4(3, 3, 3, 1))); system("pause"); return 0; ```

WaveManager: C/C++ buffer <-> PCM format wav file converter

WaveManager: C/C++ buffer <-> PCM format wav file converter === [Github repository](https://github.com/gogyzzz/cpp_butter_to_wave) ## Support buffer type short, float, double ## Support data struncture 1 dimension PCM buffer \[samplelength\] \(0ch, 1ch, 0ch, 1ch, ...\) (maximum 8channels) 2 dimension PCM buffer \[ichannel\]\[samplelength\] ## Examples ```cpp printf("************************ Test Save function ************************\n"); TYPE arr1dStatic[SAMPLESIZE]; for (int i = 0; i < SAMPLESIZE; i++) { arr1dStatic[i] = TYPE(sin(i)*0.95); } TYPE arr2dStatic[NCH][SAMPLESIZE]; for (int j = 0; j < NCH; j++) for (int i = 0; i < SAMPLESIZE; i++) { arr2dStatic[j][i] = TYPE(sin(i)*0.95); } TYPE* arr1dDyna = new TYPE[SAMPLESIZE]; for (int i = 0; i < SAMPLESIZE; i++) arr1dDyna[i] = TYPE(sin(i)*0.95); TYPE** arr2dDyna = new TYPE*[NCH]; for (int j = 0; j < NCH; j++) { arr2dDyna[j] = new TYPE[SAMPLESIZE]; for (int i = 0; i < SAMPLESIZE; i++) arr2dDyna[j][i] = TYPE(sin(i)*0.95); } wwrite("output_1d_static.wav", arr1dStatic, SAMPLESIZE); wwrite("output_2d_static.wav", arr2dStatic, SAMPLESIZE, 2); wwrite("output_1d_dynamic.wav", arr1dDyna, SAMPLESIZE); wwrite("output_2d_dynamic.wav", arr2dDyna, SAMPLESIZE, 2); printf("************************ Test Load function ************************\n"); TYPE input1dStatic[SAMPLESIZE]; int nblock = wread("output_1d_static.wav", input1dStatic); wwrite("output_1d_static_resave.wav", input1dStatic, nblock); TYPE input2dStatic[2][SAMPLESIZE]; nblock = wread("output_2d_static.wav", input2dStatic); wwrite("output_2d_static_resave.wav", input2dStatic, nblock, NCH); TYPE* input1dDyna; nblock = wread("output_1d_dynamic.wav", input1dDyna); wwrite("output_1d_dynamic_resave.wav", input1dDyna, nblock); TYPE** input2dDyna; nblock = wread("output_2d_dynamic.wav", input2dDyna); wwrite("output_2d_dynamic_resave.wav", input2dDyna, nblock, NCH); // ************************ caution ************************ // You should release memory after using pointer taken from wread. delete[] input1dDyna; for (int i = 0; i < NCH; i++) delete[] input2dDyna[i]; delete[] input2dDyna; ``` ## precautions - There are not enough safe codes. - Please do not expect "unguided" purpose.