diarize-ruby 0.3.9 → 0.3.10
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/lib/diarize/speaker.rb +2 -4
- data/lib/diarize/super_vector.rb +47 -50
- data/lib/diarize/version.rb +1 -1
- data/test/speaker_test.rb +0 -3
- data/test/super_vector_test.rb +13 -5
- data/test/version_test.rb +1 -1
- metadata +1 -1
checksums.yaml
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
---
|
2
2
|
SHA1:
|
3
|
-
metadata.gz:
|
4
|
-
data.tar.gz:
|
3
|
+
metadata.gz: 6538efc1cab1c18e656dcc50c1e2b57d8f9f38a7
|
4
|
+
data.tar.gz: d86715c3e757994d859c9847e013c5501895a5f1
|
5
5
|
SHA512:
|
6
|
-
metadata.gz:
|
7
|
-
data.tar.gz:
|
6
|
+
metadata.gz: 0762b0b955406e9c43b95e44331f518379c90a3403addcb8478bdae93746095ed060805f4e128bfd3a7640118ba821b506e467d99b3dfe3679922056eeaafac4
|
7
|
+
data.tar.gz: 0a38dcab217b7b45595d734434be3ffbdf9755c7d869bb93a49198b288b45b59d2dcdb078213d2626fc67d70bb7d955b0eeaf5982a7cb3482ea647aa9af5328d
|
data/lib/diarize/speaker.rb
CHANGED
@@ -159,8 +159,7 @@ module Diarize
|
|
159
159
|
'ws:gender' => gender,
|
160
160
|
'ws:model' => model_uri,
|
161
161
|
'ws:mean_log_likelihood' => mean_log_likelihood,
|
162
|
-
'ws:supervector_hash' => supervector.hash.to_s
|
163
|
-
'ws:sha' => supervector.sha
|
162
|
+
'ws:supervector_hash' => supervector.hash.to_s
|
164
163
|
}
|
165
164
|
end
|
166
165
|
|
@@ -169,8 +168,7 @@ module Diarize
|
|
169
168
|
'gender' => gender,
|
170
169
|
'model' => model_uri,
|
171
170
|
'mean_log_likelihood' => mean_log_likelihood,
|
172
|
-
'supervector_hash' => supervector.hash.to_s
|
173
|
-
'sha' => supervector.sha
|
171
|
+
'supervector_hash' => supervector.hash.to_s
|
174
172
|
}
|
175
173
|
end
|
176
174
|
|
data/lib/diarize/super_vector.rb
CHANGED
@@ -3,65 +3,66 @@ module Diarize
|
|
3
3
|
attr_reader :vector
|
4
4
|
|
5
5
|
def initialize(vector)
|
6
|
-
@vector = vector
|
6
|
+
@vector = vector.is_a?(Array) ? Vector.elements(vector) : vector
|
7
7
|
end
|
8
8
|
|
9
|
-
|
10
|
-
|
11
|
-
|
12
|
-
|
13
|
-
|
14
|
-
|
15
|
-
|
16
|
-
|
17
|
-
|
18
|
-
|
9
|
+
class << self
|
10
|
+
|
11
|
+
def generate_from_model(model)
|
12
|
+
# Generates a supervector from a LIUM GMM
|
13
|
+
dim = model.nb_of_components * model.components.get(0).dim
|
14
|
+
vector = Array.new(dim, 0)
|
15
|
+
model.nb_of_components.times do |k|
|
16
|
+
gaussian = model.components.get(k)
|
17
|
+
gaussian.dim.times do |i|
|
18
|
+
vector[k * gaussian.dim + i] = gaussian.mean(i)
|
19
|
+
end
|
19
20
|
end
|
21
|
+
SuperVector.new(Vector.elements(vector))
|
20
22
|
end
|
21
|
-
# SuperVector.new(vector)
|
22
|
-
SuperVector.new(Vector.elements(vector))
|
23
|
-
end
|
24
23
|
|
25
|
-
|
26
|
-
|
27
|
-
|
28
|
-
|
29
|
-
|
30
|
-
|
31
|
-
|
32
|
-
|
33
|
-
|
34
|
-
|
24
|
+
def ubm_gaussian_weights
|
25
|
+
# Returns a vector of gaussian weights, same dimension as speaker's super vectors
|
26
|
+
@@ubm_gaussian_weights ||= begin
|
27
|
+
ubm = Speaker.ubm
|
28
|
+
# weights = DoubleMatrix.new(1, ubm.supervector.dim)
|
29
|
+
weights = Array.new(ubm.supervector.dim, 0)
|
30
|
+
ubm.model.nb_of_components.times do |k|
|
31
|
+
gaussian = ubm.model.components.get(k)
|
32
|
+
gaussian.dim.times do |i|
|
33
|
+
weights[k * gaussian.dim + i] = gaussian.weight
|
34
|
+
end
|
35
35
|
end
|
36
|
+
Vector.elements(weights)
|
36
37
|
end
|
37
|
-
Vector.elements(weights)
|
38
38
|
end
|
39
|
-
end
|
40
39
|
|
41
|
-
|
42
|
-
|
43
|
-
|
44
|
-
|
45
|
-
|
46
|
-
|
47
|
-
|
48
|
-
|
49
|
-
|
50
|
-
|
40
|
+
def ubm_covariance
|
41
|
+
# Returns a vector of diagonal covariances, same dimension as speaker's super vectors
|
42
|
+
@@ubm_covariance ||= begin
|
43
|
+
ubm = Speaker.ubm
|
44
|
+
# cov = DoubleMatrix.new(1, ubm.supervector.dim)
|
45
|
+
cov = Array.new(ubm.supervector.dim)
|
46
|
+
ubm.model.nb_of_components.times do |k|
|
47
|
+
gaussian = ubm.model.components.get(k)
|
48
|
+
gaussian.dim.times do |i|
|
49
|
+
cov[k * gaussian.dim + i] = gaussian.getCovariance(i, i)
|
50
|
+
end
|
51
51
|
end
|
52
|
+
Vector.elements(cov)
|
52
53
|
end
|
53
|
-
Vector.elements(cov)
|
54
54
|
end
|
55
|
-
end
|
56
55
|
|
57
|
-
|
58
|
-
|
59
|
-
|
60
|
-
|
61
|
-
|
62
|
-
|
63
|
-
|
64
|
-
|
56
|
+
def divergence(sv1, sv2)
|
57
|
+
# ubm_gaussian_weights.mul(((sv1.vector - sv2.vector) ** 2) / ubm_covariance).sum
|
58
|
+
diff = sv1.vector - sv2.vector
|
59
|
+
square = diff.map {|el| el ** 2}
|
60
|
+
codiv = Vector.elements(square.each.with_index.inject([]) {|a,(el,ix)| a << el / ubm_covariance[ix]})
|
61
|
+
mult = ubm_gaussian_weights.each.with_index.inject([]) {|a,(el,ix)| a << el * codiv[ix]}
|
62
|
+
mult.inject(0, :+)
|
63
|
+
end
|
64
|
+
|
65
|
+
end # class
|
65
66
|
|
66
67
|
def dim
|
67
68
|
@vector.size
|
@@ -71,10 +72,6 @@ module Diarize
|
|
71
72
|
@vector.hash
|
72
73
|
end
|
73
74
|
|
74
|
-
def sha
|
75
|
-
Digest::SHA256.hexdigest(hash.to_s)
|
76
|
-
end
|
77
|
-
|
78
75
|
def to_a
|
79
76
|
@vector.to_a
|
80
77
|
end
|
data/lib/diarize/version.rb
CHANGED
data/test/speaker_test.rb
CHANGED
@@ -106,7 +106,6 @@ class SpeakerTest < Test::Unit::TestCase
|
|
106
106
|
assert_equal true, to_rdf.include?("ws:model")
|
107
107
|
assert_equal true, to_rdf.include?("ws:mean_log_likelihood")
|
108
108
|
assert_equal true, to_rdf.include?("ws:supervector_hash")
|
109
|
-
assert_equal true, to_rdf.include?("ws:sha")
|
110
109
|
assert_equal true, to_rdf.include?("https://www.example.com/model/1")
|
111
110
|
end
|
112
111
|
|
@@ -118,7 +117,6 @@ class SpeakerTest < Test::Unit::TestCase
|
|
118
117
|
assert_equal true, as_json.has_key?('model')
|
119
118
|
assert_equal true, as_json.has_key?('mean_log_likelihood')
|
120
119
|
assert_equal true, as_json.has_key?('supervector_hash')
|
121
|
-
assert_equal true, as_json.has_key?('sha')
|
122
120
|
end
|
123
121
|
|
124
122
|
def test_to_json
|
@@ -130,6 +128,5 @@ class SpeakerTest < Test::Unit::TestCase
|
|
130
128
|
assert_equal true, as_json.has_key?('model')
|
131
129
|
assert_equal true, as_json.has_key?('mean_log_likelihood')
|
132
130
|
assert_equal true, as_json.has_key?('supervector_hash')
|
133
|
-
assert_equal true, as_json.has_key?('sha')
|
134
131
|
end
|
135
132
|
end
|
data/test/super_vector_test.rb
CHANGED
@@ -15,16 +15,24 @@ class SuperVectorTest < Test::Unit::TestCase
|
|
15
15
|
end
|
16
16
|
end
|
17
17
|
|
18
|
-
def
|
18
|
+
def test_initialize_from_supervector
|
19
19
|
model = Diarize::Speaker.load_model(File.join(File.dirname(__FILE__), 'data', 'speaker1.gmm'))
|
20
|
-
|
21
|
-
|
20
|
+
sv1 = Diarize::SuperVector.generate_from_model(model)
|
21
|
+
sv2 = Diarize::SuperVector.new(sv1)
|
22
|
+
assert_equal sv1.hash, sv2.hash
|
23
|
+
end
|
24
|
+
|
25
|
+
def test_initialize_from_array
|
26
|
+
model = Diarize::Speaker.load_model(File.join(File.dirname(__FILE__), 'data', 'speaker1.gmm'))
|
27
|
+
sv1 = Diarize::SuperVector.generate_from_model(model)
|
28
|
+
sv2 = Diarize::SuperVector.new(sv1.to_a)
|
29
|
+
assert_equal sv1.hash, sv2.hash
|
22
30
|
end
|
23
31
|
|
24
|
-
def
|
32
|
+
def test_hash
|
25
33
|
model = Diarize::Speaker.load_model(File.join(File.dirname(__FILE__), 'data', 'speaker1.gmm'))
|
26
34
|
sv = Diarize::SuperVector.generate_from_model(model)
|
27
|
-
assert_equal
|
35
|
+
assert_equal sv.vector.hash, sv.hash
|
28
36
|
end
|
29
37
|
|
30
38
|
def test_to_a
|
data/test/version_test.rb
CHANGED