-
Notifications
You must be signed in to change notification settings - Fork 13
/
README
125 lines (86 loc) · 5.57 KB
/
README
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
#######################################################################
########### VOICE ANALYSIS TOOLKIT ###################################
######################################################################
This software has been coded by John Kane at the Phonetics and Speech
Laboratory in Trinity College Dublin, 2009-2013. This work is supported
by the Science Foundation Ireland, Grant 07/CE/I1142 (Centre for Next
Generation Localisation, www.cngl.ie) and Grant 09/IN.1/I2631 (FASTNET).
The toolkit contains a range of matlab files for glottal source and
voice quality analysis. For most of the algorithms Matlab versions
since 2009 be suitable. However, for the CreakyDetection_CompleteDetection.m function only Matlab versions since Matlab R2011a (and including the neural network toolbox) will be able to run it. Note that the creak detection algorithm used here was developed jointly by John Kane (Phonetics and Speech Laboratory in Trinity College Dublin) and Thomas Drugman (University of Mons, Belgium) and that same algorithm is also available within the GLOAT toolkit.
GETTING STARTED
If requiring the use of the LF mode function the lf_Area_newton.c in the general_fcns/
directory must be mex-ed, e.g., use this command:
mex lf_Area_newton.c
in the general_fcns/ directory.
Note however there mex-ed versions are already included for Linux (32-bit), Mac (64-bit) and Windows (32-bit and 64-bit).
Note also that the SRH f0/VUV tracking algorithm from the GLOAT toolkit
(Thomas Drugman, University of Mons) is used with this toolkit. The GLOAT
toolkit can be downloaded from here:
GLOAT - http://tcts.fpms.ac.be/~drugman/Toolbox/GLOAT.zip
and the publication:
T.Drugman, A.Alwan, Joint Robust Voicing Detection and Pitch Estimation Based on Residual Harmonics, Interspeech11, Firenze, Italy, 2011
To see details of the usage of each of the methods, simply use the
help command in matlab, e.g.,
help SE_VQ
TESTING
To do a test-run of the software on an included ARCTIC utterance,
use the command:
[x,fs]=wavread('arctic_a0007.wav');
test_Voice_Analysis_Toolkit(x,fs)
FEEDBACK
Note that the code here has been developed and tested in a UNIX based
environment. None of the directory handling has been hardcoded and hence
there should not be problems running this on a windows machine. Also,
the code has been developed for Matlab 2011b. There may be unforeseen
problems with previous (and potentially future) versions. If you have
any difficulties or issues with the code please email me: [email protected]
NOVEL ALGORITHMS - BRIEF DESCRIPTION
SE_VQ - Algorithm for detecting glottal closure instants. This is a further
further development of the SEDREAMS algorithm (Drugman et al 2012)
and is designed to improve selection of GCI candidates through the
use of a dynamic programming algorithm. It also involves a
post-processing step to remove false positives in creaky
voice regions.
dyProg_LF (COMING SOON!!) - Algorithm for fitting LF model pulses to an estimated glottal
flow derivative waveform. The method involves an exhaustive search
process using Rd, a dynamic programming algorithm to choose the
optimal path of Rd values and a subsequent optimisation algorithm
in order to refine the fit by varying all three R-parameters
MDQ - The maxima dispersion quotient (MDQ) is used for discriminating
breathy to tense voice based on wavelet-based decomposition of the
LP-residual signal. Dispersion is measured in the vicinity of the GCI
creak_detect - An algorithm for detecting creaky voice regions from speech
signals. The detection is based on a combination of new and existing
acoustic features relevant to creaky voice which are used as input
features to a artificial neural networks based classifier, which has been trained
on a wide range of speech. Note that the development of this method
has been done in collaboration with Thomas Drugman, University of Mons,
Belgium.
REFERENCES
Please refer to the relevant references below when using any of these
algorithms in published studies.
Kane, J., Gobl, C., (2013) `Evaluation of glottal closure instant
detection in a range of voice qualities', Speech Communication
55(2), pp. 295-314.
Kane, J., Gobl, C. (2013) ``Automating manual user strategies for
precise voice source analysis'', Speech Communication 55(3), pp.
397-414.
Kane, J., Yanushevskaya, I., Ni Chasaide, A., Gobl, C., (2012) Exploiting time
and frequency domain measures for precise voice source parameterisation,
Proceedings of Speech Prosody.
Kane, J., Gobl, C., (2013)``Wavelet maxima dispersion for breathy to tense
voice discrimination'', IEEE Trans. Audio Speech & Language
Processing, 21(6), pp. 1170-1179.
Kane, J., Gobl, C., (2011) `Identifying regions of non-modal phonation using
features of the wavelet tranform', Proceedings of Interspeech 2011.
Drugman, T., Kane, J., Gobl, C., `Automatic Analysis of Creaky
Excitation Patterns', Submitted to Computer Speech and
Language.
Kane, J., Drugman, T., Gobl, C., (2013) `Improved automatic
detection of creak', 27(4), pp. 1028-1047, Computer Speech and Language.
Drugman, T., Kane, J., Gobl, C. (2012) Resonator-based creaky voice detection,
Proceedings of Interspeech.
Drugman, T., Kane, J., Gobl, C. (2012) Modeling the creaky excitation for
parametric speech synthesis, Proceedings of Interspeech.
######################################################################