| 1 |
Disclaimer: this is a *sanity test* only, and is *not* to be regarded as a valid test of the VoxForge Acoustic Models! |
|---|
| 2 |
* the audio files used for testing include only one voice, the Acoustic Models were also trained using that same voice, so the results will look better; |
|---|
| 3 |
* there are only 50 audio samples in the test database - not enough for a good test; |
|---|
| 4 |
* the VoxForge Acoustic Models are still alpha with respect to Speaker Independent Speech Recognition, so please donate some speech to VoxForge, |
|---|
| 5 |
thanks, |
|---|
| 6 |
Ken |
|---|
| 7 |
|
|---|
| 8 |
Testing Acoustic Models created in: /data/svn-mirror-copy/Nightly_Builds/AcousticModel-2007-01-25 |
|---|
| 9 |
|
|---|
| 10 |
HTK 16kHz_16bit |
|---|
| 11 |
--------------- |
|---|
| 12 |
word insertion penalty: 0.0 |
|---|
| 13 |
grammar scale factor: 1.0 |
|---|
| 14 |
====================== Results Analysis ======================= |
|---|
| 15 |
Date: Thu Jan 25 13:48:15 2007 |
|---|
| 16 |
Ref : testref.mlf |
|---|
| 17 |
Rec : recout.mlf |
|---|
| 18 |
------------------------ Overall Results -------------------------- |
|---|
| 19 |
SENT: %Correct=22.00 [H=11, S=39, N=50] |
|---|
| 20 |
WORD: %Corr=86.24, Acc=44.44 [H=163, D=1, S=25, I=79, N=189] |
|---|
| 21 |
=================================================================== |
|---|
| 22 |
|
|---|
| 23 |
Julian 16kHz_16bit |
|---|
| 24 |
------------------ |
|---|
| 25 |
word insertion penalty |
|---|
| 26 |
first pass (-penalty1):0.5 |
|---|
| 27 |
second pass (-penalty2):100.0 |
|---|
| 28 |
transition penalty:-55.0 (for short-term inter-word pauses between words (-iwsppenalty)) |
|---|
| 29 |
====================== Results Analysis ======================= |
|---|
| 30 |
Date: Thu Jan 25 13:48:19 2007 |
|---|
| 31 |
Ref : testref.mlf |
|---|
| 32 |
Rec : julianProcessed |
|---|
| 33 |
------------------------ Overall Results -------------------------- |
|---|
| 34 |
SENT: %Correct=88.00 [H=44, S=6, N=50] |
|---|
| 35 |
WORD: %Corr=97.35, Acc=96.83 [H=184, D=2, S=3, I=1, N=189] |
|---|
| 36 |
=================================================================== |
|---|
| 37 |
|
|---|
| 38 |
HTK 8kHz_16bit |
|---|
| 39 |
--------------- |
|---|
| 40 |
word insertion penalty: 10.0 |
|---|
| 41 |
grammar scale factor: 5.0 |
|---|
| 42 |
====================== Results Analysis ======================= |
|---|
| 43 |
Date: Thu Jan 25 13:48:21 2007 |
|---|
| 44 |
Ref : testref.mlf |
|---|
| 45 |
Rec : recout.mlf |
|---|
| 46 |
------------------------ Overall Results -------------------------- |
|---|
| 47 |
SENT: %Correct=52.00 [H=26, S=24, N=50] |
|---|
| 48 |
WORD: %Corr=83.07, Acc=65.08 [H=157, D=1, S=31, I=34, N=189] |
|---|
| 49 |
=================================================================== |
|---|
| 50 |
|
|---|
| 51 |
Julian 8kHz_16bit |
|---|
| 52 |
------------------ |
|---|
| 53 |
word insertion penalty |
|---|
| 54 |
first pass (-penalty1):50.0 |
|---|
| 55 |
second pass (-penalty2):100.0 |
|---|
| 56 |
transition penalty::-55.0 (for short-term inter-word pauses between words (-iwsppenalty)) |
|---|
| 57 |
====================== Results Analysis ======================= |
|---|
| 58 |
Date: Thu Jan 25 13:48:23 2007 |
|---|
| 59 |
Ref : testref.mlf |
|---|
| 60 |
Rec : julianProcessed |
|---|
| 61 |
------------------------ Overall Results -------------------------- |
|---|
| 62 |
SENT: %Correct=84.00 [H=42, S=8, N=50] |
|---|
| 63 |
WORD: %Corr=95.24, Acc=91.53 [H=180, D=1, S=8, I=7, N=189] |
|---|
| 64 |
=================================================================== |
|---|
| 65 |
|
|---|
| 66 |
Notes: |
|---|
| 67 |
|
|---|
| 68 |
* the line starting with SENT gives the percentage of sentences that were recognized correctly, out of N sentences in total. |
|---|
| 69 |
* the line starting with WORD gives the percentage of words that were recognized correctly, out of N words in total |
|---|
| 70 |
However, since HTK or Julius erroneously 'added' words that are not in the audio file (i.e. insertion errors) they usually get a lower percentage accuracy rating. |
|---|
| 71 |
* Count definitions: |
|---|
| 72 |
o D - Deletion Error |
|---|
| 73 |
o S - Substitution Error |
|---|
| 74 |
o I - Insertion Error |
|---|
| 75 |
|
|---|
| 76 |
|
|---|
| 77 |
|
|---|
| 78 |
================================================================================================================ |
|---|
| 79 |
For comparison purposes, see below for the same Tests on the most current release of the VoxForge Acoustic Models: |
|---|
| 80 |
(/data/svn-mirror-copy/Tags/Releases/0_1_1-build726) |
|---|
| 81 |
================================================================================================================ |
|---|
| 82 |
HTK 16kHz_16bit |
|---|
| 83 |
--------------- |
|---|
| 84 |
word insertion penalty: 0.0 |
|---|
| 85 |
grammar scale factor: 1.0 |
|---|
| 86 |
====================== Results Analysis ======================= |
|---|
| 87 |
Date: Thu Jan 25 13:48:30 2007 |
|---|
| 88 |
Ref : testref.mlf |
|---|
| 89 |
Rec : recout.mlf |
|---|
| 90 |
------------------------ Overall Results -------------------------- |
|---|
| 91 |
SENT: %Correct=22.00 [H=11, S=39, N=50] |
|---|
| 92 |
WORD: %Corr=88.36, Acc=50.79 [H=167, D=1, S=21, I=71, N=189] |
|---|
| 93 |
=================================================================== |
|---|
| 94 |
|
|---|
| 95 |
Julian 16kHz_16bit |
|---|
| 96 |
------------------ |
|---|
| 97 |
word insertion penalty |
|---|
| 98 |
first pass (-penalty1):0.5 |
|---|
| 99 |
second pass (-penalty2):100.0 |
|---|
| 100 |
transition penalty:-55.0 (for short-term inter-word pauses between words (-iwsppenalty)) |
|---|
| 101 |
====================== Results Analysis ======================= |
|---|
| 102 |
Date: Thu Jan 25 13:48:33 2007 |
|---|
| 103 |
Ref : testref.mlf |
|---|
| 104 |
Rec : julianProcessed |
|---|
| 105 |
------------------------ Overall Results -------------------------- |
|---|
| 106 |
SENT: %Correct=86.00 [H=43, S=7, N=50] |
|---|
| 107 |
WORD: %Corr=96.83, Acc=96.30 [H=183, D=2, S=4, I=1, N=189] |
|---|
| 108 |
=================================================================== |
|---|
| 109 |
|
|---|
| 110 |
HTK 8kHz_16bit |
|---|
| 111 |
--------------- |
|---|
| 112 |
word insertion penalty: 10.0 |
|---|
| 113 |
grammar scale factor: 5.0 |
|---|
| 114 |
====================== Results Analysis ======================= |
|---|
| 115 |
Date: Thu Jan 25 13:48:40 2007 |
|---|
| 116 |
Ref : testref.mlf |
|---|
| 117 |
Rec : recout.mlf |
|---|
| 118 |
------------------------ Overall Results -------------------------- |
|---|
| 119 |
SENT: %Correct=56.00 [H=28, S=22, N=50] |
|---|
| 120 |
WORD: %Corr=82.01, Acc=65.08 [H=155, D=2, S=32, I=32, N=189] |
|---|
| 121 |
=================================================================== |
|---|
| 122 |
|
|---|
| 123 |
Julian 8kHz_16bit |
|---|
| 124 |
------------------ |
|---|
| 125 |
word insertion penalty |
|---|
| 126 |
first pass (-penalty1):50.0 |
|---|
| 127 |
second pass (-penalty2):100.0 |
|---|
| 128 |
transition penalty::-55.0 (for short-term inter-word pauses between words (-iwsppenalty)) |
|---|
| 129 |
====================== Results Analysis ======================= |
|---|
| 130 |
Date: Thu Jan 25 13:48:41 2007 |
|---|
| 131 |
Ref : testref.mlf |
|---|
| 132 |
Rec : julianProcessed |
|---|
| 133 |
------------------------ Overall Results -------------------------- |
|---|
| 134 |
SENT: %Correct=86.00 [H=43, S=7, N=50] |
|---|
| 135 |
WORD: %Corr=95.77, Acc=92.59 [H=181, D=2, S=6, I=6, N=189] |
|---|
| 136 |
=================================================================== |
|---|
| 137 |
|
|---|
| 138 |
Notes: |
|---|
| 139 |
|
|---|
| 140 |
* the line starting with SENT gives the percentage of sentences that were recognized correctly, out of N sentences in total. |
|---|
| 141 |
* the line starting with WORD gives the percentage of words that were recognized correctly, out of N words in total |
|---|
| 142 |
However, since HTK or Julius erroneously 'added' words that are not in the audio file (i.e. insertion errors) they usually get a lower percentage accuracy rating. |
|---|
| 143 |
* Count definitions: |
|---|
| 144 |
o D - Deletion Error |
|---|
| 145 |
o S - Substitution Error |
|---|
| 146 |
o I - Insertion Error |
|---|