UnderdeterminedSeparationofInstaneousSpeechMixture
所属分类:matlab编程
开发工具:matlab
文件大小:222KB
下载次数:255
上传日期:2010-05-28 09:16:12
上 传 者:
lpbjob1217
说明: 详细描述了关于几个同时说话人语音分离的情况,其中包括了原代码和测试语音,测试结果表明分离效果很好,能够区分不同人说话的语音,明显提高语音识别率。其中的参考论文在readme中。
(A detailed description about several while speaking voice separate conditions, including the original code and test speech, test results show that the effect is very good separation between different people speak of speech recognition, obviously improved pronunciation. One of the reference in the readme.)
文件列表:
stereomix.wav (200044, 2006-01-20)
jpfe.wav (100044, 2005-02-03)
icaML.m (15508, 2005-04-26)
polara.m (5164, 2005-05-12)
flotmatrix.m (2726, 2005-10-12)
calcA.m (144, 2006-01-18)
invspecgram.m (1007, 2006-01-19)
normmix.m (198, 2006-01-19)
evalICAdB.m (1746, 2006-01-19)
oneortwo_cond.m (496, 2006-01-19)
enerstop.m (919, 2006-01-19)
colorimask.m (1566, 2006-01-19)
sg.m (1031, 2006-01-20)
idealmask.m (957, 2006-01-20)
applymasks.m (3448, 2006-01-20)
getfinalmask.m (875, 2006-01-20)
multisigcheck.m (1718, 2006-01-20)
nosigcorr.m (2128, 2006-01-20)
getremainingmask.m (883, 2006-01-20)
calcELNR.m (4259, 2006-01-20)
comparemasks.m (2793, 2006-01-20)
This is the source code for underdetermined separation of instaneous speech mixtures.
The algorithm is described in
[1] Michael Syskind Pedersen, DeLiang Wang, Jan Larsen and Ulrik Kjems:
Two-microphone Separation of Speech Mixtures, 2006, Submitted for publication.
See also
[2] Michael Syskind Pedersen, DeLiang Wang, Jan Larsen and Ulrik Kjems, Overcomplete Blind Source Separation by
Combining ICA and Binary Time-Frequency Masking, IEEE International workshop on Machine
Learning for Signal Processing, pp. 15-20, 2005
All files should be in the same directory.
The algorithm is run by calling main.m.
A number of parameters can be specified in main.m:
N Number of sources in mixture
NFFT DFT length
winnumber Selects window function
k Window length is NFFT/k
noverlapfactor Overlap between consecutive windows
th Mask threshold?
TC1 Merge finalstereo signals if correlation is above TC1
TC2 Merge finalstereo and enerstereo if correlation is above TC2
stopthresholdini One source if condition number is above this value
thepow tau_E (see [1])
A sound demo is available online: http://www.imm.dtu.dk/pubdb/p.php?4400
2006-01-20
近期下载者:
相关文件:
收藏者: