matlab语音识别的一些基础代码库

function [enhanced_speech] = spendred(input_speech,fs,algo_params) % SPENDRED Speech Enhancement and Dereverberation by Doire % % Usage : (1) [enhanced_speech] = spendred(corrupted_speech,fs) % performs enhancement in order to denoise and dereverberate the input speech file % (2) [enhanced_speech] = spendred(corrupted_speech,fs,algo_params) % performs enhancement using the custom parameters specified in 'algo_params' % % % Inputs: % input_speech Noisy and reverberant input speech signal (single-channel) % fs Sample frequency in Hz % algo_params algorithm parameters [optional] % % % Outputs: % enhanced_speech Enhanced output speech file % % Algorithm Parameters: % The following parameters are defined in 'algo_params'. Default values are shown below. % % algo_params.of=6 % Overlap factor [6] % algo_params.ti=0.005 % Desired frame increment in seconds [0.005] % algo_params.ri=1 % Set to 1 to round ti to the nearest power of 2 samples [1] % algo_params.sg=1 % Type of spectral subtraction gain to apply : 1=Wiener Gain, 2=Power Spectral Subtraction, 3=MMSE speech estimate [1] % algo_params.sc=0.95 % Smoothing constant for computation of the spectral gain [0.95] % algo_params.sf=1e-5; % Floor for the spectral gain [1e-5] % algo_params.os=2; % Interference Over-subtraction factor [2] % algo_params.cl=6; % Number of HMM states to use (minimum is 2 - maximum is 6) [6] % algo_params.ds=1; % Way of computing posterior distributions : 1 = max track , 2 = weighted sum of tracks % algo_params.mo=1; % Mode : 'fast' = 1 or 'slow' = 0 [1] % % % References: % % [1] C.S.J. Doire, M. Brookes, P.A. Naylor, C.M. Hicks, D. Betts, % M.A. Dmour, S.H. Jensen, "SINGLE-CHANNEL ONLINE ENHANCEMENT OF SPEECH % CORRUPTED BY REVERBERATION AND NOISE", IEEE Trans. Audio, Speech, % Language Processing, accepted for publication: 1�16, 2017. % % Author : Clement Doire % clement.doire11@imperial.ac.uk % % % Revision History: % % 0.5 - 16 Feb 2016 - First version, based on my PhD thesis and first draft of [1] % 0.9 - 13 Mar 2016 - Several tweaks + improved overall performance % 1.0 - 01 Jul 2016 - Base version used to generate results in the paper % % Comments: % - By default the algorithm is in 'fast' mode, which fails in rare % cases due to numerical errors. This can be overcome by using a Square Root implementation, % which is done via the 'slow' mode option algo_params.mo=0. % % Copyright (C) Clement Doire 2016 % Version: $Id: spendred.m 9214 2016-12-17 17:13:24Z dmb $ % % VOICEBOX is a MATLAB toolbox for speech processing. % Home page: http://www.ee.ic.ac.uk/hp/staff/dmb/voicebox/voicebox.html % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % This program is free software; you can redistribute it and/or modify % it under the terms of the GNU General Public License as published by % the Free Software Foundation; either version 2 of the License, or % (at your option) any later version. % % This program is distributed in the hope that it will be useful, % but WITHOUT ANY WARRANTY; without even the implied warranty of % MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the % GNU General Public License for more details. % % You can obtain a copy of the GNU General Public License from % http://www.gnu.org/copyleft/gpl.html or by writing to % Free Software Foundation, Inc.,675 Mass Ave, Cambridge, MA 02139, USA. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%% TO DO : Deal with other sampling freqs more gracefully %%%% if (fs ~= 16000) input_speech = resample(mean(input_speech,2),16000,fs); fs_ori = fs; fs = 16000; end %%%% %%%% if numel(input_speech)>length(input_speech) error('Input speech signal must be a vector, not a matrix !'); end %Default algorithm constants Csts.of=6; % Overlap factor = (fft length)/(frame increment) [6] Csts.ti=5e-3; % Desired frame increment [0.005] Csts.ri=1; % Round ni to the nearest power of 2 [1] Csts.sg=1; % Type of spectral subtraction gain to apply : 1=Wiener Gain, 2=Power Spectral Subtraction, 3=MMSE speech estimate [1] Csts.sc=0.95; % Smoothing constant for computation of the spectral gain [0.95] Csts.sf=1e-5; % Floor for the spectral gain [1e-5] Csts.os=2; % Interference Over-subtraction factor [2] Csts.cl=6; % Number of HMM states to use (minimum is 2 - maximum is 6) [6] Csts.ds=1; % Way of computing posterior distributions : 1 = max track , 2 = weighted sum of tracks Csts.mo=1; % Mode : 'fast' = 1 or 'slow' = 0 [1] if nargin>=3 && ~isempty(algo_params) qqn=fieldnames(Csts); for i=1:length(qqn) if isfield(algo_params,qqn{i}) Csts.(qqn{i})=algo_params.(qqn{i}); end end end %-------------------------------------------------------------------------% %-------------------------------------------------------------------------% %load the dictionary mStates = [-28.4861773669176,-19.9241782718819,-16.4426351907036,... -13.2371952081660,-9.08277302468020,-7.94392920991245,-10.1807698459865,... -13.0242640127927,-15.0862478706129,-17.1550659704425,-20.5281316935197,-24.6762435506711,-28.8594807710477,-32.5786959842695,-34.7893823914205,... -36.3359261522268,-36.8192673337429,-36.4363892290271,-38.0294065203303,-42.0503710814368,-45.5963723705087,-48.2040932769337,-49.5383787887783,... -49.1147104408592,-48.8102903134417;-55.6639981588596,-61.5127504297246,-61.7239465777425,-62.8243042292255,-63.6413685588512,-63.4686731115175,... -63.7758454966487,-63.6874780526371,-63.8179939306919,-63.9503258211402,-63.7966111988872,-64.0463832623785,-64.0818520961556,-64.1192957981112,... -63.8411944409695,-63.6617787810601,-63.5628614840156,-63.1631597694168,-62.5434531164831,-61.9566668567832,-61.5323786575963,-61.3910567729765,... -61.1150168405862,-60.5161965622102,-59.9977551298916;-28.6934939143922,-19.2542393025089,-15.1143438538347,-12.7293198179762,-12.1005903838332,... -16.5017610503493,-21.3970568254212,-25.1613543930201,-28.8424911990220,-31.5618412114192,-32.1889756387523,-29.6313245488876,-26.0094143727967,... -23.9888695941115,-23.0689428325671,-23.6462241160502,-26.4296918344984,-27.1090165222226,-27.8756237542369,-31.1075250195298,-36.3923980939477,... -40.8889238387664,-43.4857021995535,-43.3831690409702,-42.9845505669993;-45.3678779268938,-43.6673845369666,-41.8441402268909,-39.9311165545071,... -37.8535557910413,-36.8469454100004,-36.6393657864992,-36.5362981350264,-36.9937748890843,-37.4953438142185,-37.0039280540463,-35.1723035151364,... -33.6115205373159,-33.4564666785186,-32.0578128508397,-29.7739710941898,-28.4547720864261,-25.9299736823261,-23.4486089524545,-21.9613450596309,... -21.4217191910235,-22.0225885633808,-23.1188663976374,-23.7853960376589,-24.1662738392603;-34.1432887176235,-26.3187620722736,-23.6590932454401,... -22.8473441058075,-23.1748279271661,-26.4411088266502,-29.2521006269876,-31.1772170063613,-33.6716054384278,-36.4153183945587,-38.6110013517002,... -39.9143397342699,-41.0618568585801,-41.5304011208901,-41.0824724701060,-42.0444890310335,-44.6508339686914,-45.7999391511713,-45.7777619091791,... -46.4288731995784,-48.2850680725583,-50.4157570788352,-51.6879601727512,-51.5283308152838,-51.3663633824326;-27.8144244788325,-17.3617936140541,... -13.3123479510278,-11.0494334946260,-6.50571075754786,-3.54362892667645,-4.20504911233702,-8.05475391358820,-11.8924915327701,-14.0090223475965,... -14.4546676449037,-14.2927721880105,-15.9448322180079,-18.7341480479014,-19.59665