sphinx2-test
Spoiler:
Quote:
Run CMU Sphinx2 in Batch mode to decode an example utterance.
/usr/bin/sphinx2-continuous \
-verbose 9 \
-adcin TRUE \
-adcext 16k \
-ctlfn /usr/share/sphinx2/model/lm/turtle/turtle.ctl \
-ctloffset 0 \
-ctlcount 100000000 \
-datadir /usr/share/sphinx2/model/lm/turtle \
-agcmax TRUE \
-langwt 6.5 \
-fwdflatlw 8.5 \
-rescorelw 9.5 \
-ugwt 0.5 \
-fillpen 1e-10 \
-silpen 0.005 \
-inspen 0.65 \
-top 1 \
-topsenfrm 3 \
-topsenthresh \
-70000 \
-beam 2e-06 \
-npbeam 2e-06 \
-lpbeam 2e-05 \
-lponlybeam 0.0005 \
-nwbeam 0.0005 \
-fwdflat FALSE \
-fwdflatbeam 1e-08 \
-fwdflatnwbeam 0.0003 \
-bestpath TRUE \
-kbdumpdir /usr/share/sphinx2/model/lm/turtle \
-lmfn /usr/share/sphinx2/model/lm/turtle/turtle.lm \
-dictfn /usr/share/sphinx2/model/lm/turtle/turtle.dic \
-ndictfn /usr/share/sphinx2/model/hmm/6k/noisedict \
-phnfn /usr/share/sphinx2/model/hmm/6k/phone \
-mapfn /usr/share/sphinx2/model/hmm/6k/map \
-hmmdir /usr/share/sphinx2/model/hmm/6k \
-hmmdirlist /usr/share/sphinx2/model/hmm/6k \
-8bsen TRUE \
-sendumpfn /usr/share/sphinx2/model/hmm/6k/sendump \
-cbdir /usr/share/sphinx2/model/hmm/6k
kindle
Sat Mar 23 15:24:20 GMT 2013
INFO: fbs_main.c(1266): libfbs/main COMPILED ON: Apr 15 2008, AT: 18:38:14
INFO: kb_main.c(601): Reading phone file [/usr/share/sphinx2/model/hmm/6k/phone]
INFO: phone.c(387): Using 0 phonological mappings
INFO: kb_main.c(610): Reading map file [/usr/share/sphinx2/model/hmm/6k/map]
INFO: hmm_tied_r.c(1339): Read Map: 125716 phones map to 23355 unique senone sequences
INFO: kb_main.c(648): Reading dict file [/usr/share/sphinx2/model/lm/turtle/turtle.dic]
INFO: dict.c(473): dict_load: first line of /usr/share/sphinx2/model/lm/turtle/turtle.dic was A, expecting '!'
INFO: dict.c(475): dict_load: will assume first line contains a word
INFO: dict.c(639): 111 = words in file [/usr/share/sphinx2/model/lm/turtle/turtle.dic]
INFO: dict.c(473): dict_load: first line of /usr/share/sphinx2/model/hmm/6k/noisedict was ++BREATH++, expecting '!'
INFO: dict.c(475): dict_load: will assume first line contains a word
INFO: dict.c(639): 7 = words in file [/usr/share/sphinx2/model/hmm/6k/noisedict]
INFO: dict.c(406): LEFT CONTEXT TABLES
INFO: dict.c(1083): Entry Context table contains
76 entries
INFO: dict.c(1084): 3876 possible cross word triphones.
INFO: dict.c(1120): 3268 triphones
608 pseudo diphones
0 uniphones
INFO: dict.c(1146): Exit Context table contains
76 entries
INFO: dict.c(1147): 3876 possible cross word triphones.
INFO: dict.c(1210): 3268 triphones
608 pseudo diphones
0 uniphones
INFO: dict.c(1212): 1605 right context entries
INFO: dict.c(1213): 21 ave entries per exit context
INFO: dict.c(410): RIGHT CONTEXT TABLES
INFO: dict.c(1083): Entry Context table contains
65 entries
INFO: dict.c(1084): 3315 possible cross word triphones.
INFO: dict.c(1120): 2730 triphones
585 pseudo diphones
0 uniphones
INFO: dict.c(1146): Exit Context table contains
65 entries
INFO: dict.c(1147): 3315 possible cross word triphones.
INFO: dict.c(1210): 2730 triphones
585 pseudo diphones
0 uniphones
INFO: dict.c(1212): 1625 right context entries
INFO: dict.c(1213): 25 ave entries per exit context
INFO: dict.c(414): 0 unique triphones were mapped to ci phones
INFO: lm_3g.c(864): Reading LM file /usr/share/sphinx2/model/lm/turtle/turtle.lm (name "")
INFO: lm_3g.c(894): ngrams 1=91, 2=212, 3=177
INFO: lm_3g.c(901): 121 words in dictionary
INFO: lm_3g.c(1416): Looking for precompiled LM dump file /usr/share/sphinx2/model/lm/turtle/turtle.lm.DMP
INFO: lm_3g.c(1419): Precompiled file not found; continue with LM file
INFO: lm_3g.c(562): Reading unigrams
INFO: lm_3g.c(943): 91 = #unigrams created
INFO: lm_3g.c(606): Reading bigrams
INFO: lm_3g.c(955):
212 = #bigrams created
INFO: lm_3g.c(956): 20 = #prob2 entries
INFO: lm_3g.c(963): 12 = #bo_wt2 entries
INFO: lm_3g.c(684): Reading trigrams
INFO: lm_3g.c(972):
177 = #trigrams created
INFO: lm_3g.c(973): 12 = #prob3 entries
INFO: lm_3g.c(1070): bo_wt(</s>) changed from -0.3010 to -99.0000
INFO: lm_3g.c(1076): prob(<s>) changed from -0.9129 to -99.0000
INFO: lm_3g.c(1722): 6.50 = Language Weight
INFO: lm_3g.c(1723): 0.50 = Unigram Weight
INFO: lm_3g.c(1724): -4308 = LOG (Insertion Penalty (0.65))
INFO: lm_3g.c(1218): LM("") added
INFO: lm_3g.c(1148): Adding 0 initial OOV words to LM
INFO: hmm_tied_r.c(244): Loading HMMs from dump file /usr/share/sphinx2/model/hmm/6k/sendump
INFO: hmm_tied_r.c(254): BEGIN FILE FORMAT DESCRIPTION
INFO: fbs_main.c(1287): Not using S3 continuous models; initializing SCVQ module
INFO: sc_vq.c(545): SCVQInit: top 1, 6255 models, 1 dist, 0.000100 var floor.
INFO: cep_rw.c(140): Byte reversing /usr/share/sphinx2/model/hmm/6k/cep.256.vec
INFO: cep_rw.c(140): Byte reversing /usr/share/sphinx2/model/hmm/6k/cep.256.var
INFO: sc_vq.c(1511): sc_vq.c(1511): Quantizing senone PDFs to 8 bits
INFO: cep_rw.c(140): Byte reversing /usr/share/sphinx2/model/hmm/6k/d2cep.256.vec
INFO: cep_rw.c(140): Byte reversing /usr/share/sphinx2/model/hmm/6k/d2cep.256.var
INFO: sc_vq.c(1511): sc_vq.c(1511): Quantizing senone PDFs to 8 bits
INFO: cep_rw.c(140): Byte reversing /usr/share/sphinx2/model/hmm/6k/p3cep.256.vec
INFO: cep_rw.c(140): Byte reversing /usr/share/sphinx2/model/hmm/6k/p3cep.256.var
INFO: sc_vq.c(1511): sc_vq.c(1511): Quantizing senone PDFs to 8 bits
INFO: cep_rw.c(140): Byte reversing /usr/share/sphinx2/model/hmm/6k/xcep.256.vec
INFO: cep_rw.c(140): Byte reversing /usr/share/sphinx2/model/hmm/6k/xcep.256.var
INFO: sc_vq.c(1511): sc_vq.c(1511): Quantizing senone PDFs to 8 bits
INFO: search.c(3469): Estimating maximal search tree
INFO: search.c(3580): max nonroot chan increased to 337
INFO: search.c(3590): 73 root, 209 non-root channels, 13 single-phone words
INFO: search.c(2020): topsen-window = 3, threshold = -70000
INFO: search.c(2889): -1049840 = beam width
INFO: search.c(2898): -608104 = new word beam width
INFO: search.c(2916): -1049840 = new phone beam width
INFO: search.c(2925): -865624 = last phone beam width
INFO: search.c(2907): -608104 = Last phone alone beam width
INFO: search.c(2969): -52986 = LOG (Silence Word Penalty) + LOG (Phone Penalty)
INFO: search.c(2982): -230270 = LOG (Filler Word Penalty) + LOG (Phone Penalty)
INFO: search.c(2960): 0 = newword penalty
INFO: search.c(2993): LW = fwdtree: 6.5, fwdflat: 8.5, bestpath: 9.5
INFO: search.c(2995): Log2(AscrScale) = 0
INFO: search.c(4020): Flat-pass bw = 1.0e-08 (-1473728), nwbw = 3.0e-04 (-648968)
INFO: fbs_main.c(2299): Will not use double bandwidth in mel filter
INFO: live_norm.c(123): mean[0]= 8.00, mean[1..12]= 0.0
INFO: fsg_search.c(226): FSG(beam: -1049840, pbeam: -1049840, wbeam: -608104; wip: -28002, pip: 0)
INFO: uttproc.c(1884): CMN: Based on current utterance
INFO: uttproc.c(1908): AGC: MAX based on current utterance
INFO: search.c(3467): Creating search tree
INFO: search.c(3590): 73 root, 209 non-root channels, 13 single-phone words
INFO: uttproc.c(1743): LM= ""
INFO: search.c(2110): startword= <s> (id= 112)
INFO: fbs_main.c(1413): libfbs/main COMPILED ON: Apr 15 2008, AT: 18:38:14
INFO: fbs_main.c(1550):
Utterance: goforward
INFO: uttproc.c(1186): Batchmode
INFO: uttproc.c(1382): Samples histogram (goforward) (4/8/16/30/32K): 100.0%(80872) 0.0%(0) 0.0%(0) 0.0%(0) 0.0%(0); max: 1035
6.378 = AGC MAX
SFrm Efrm AScr/Frm AScr LScr BSDiff LatDen PhPerp Word (FWDTREE) (goforward)
---------------------------------------------------------------------
0 56 -187845 -10707215 0 -180727 1 0.29 <s>
57 62 -234546 -1407281 -52986 -207872 2 1.25 SIL
63 76 -207852 -2909941 -167172 -141750 8 1.19 GO
77 85 -213946 -1925516 -52986 -164181 2 1.26 SIL
86 117 -235026 -7520843 -94430 -167232 2 0.89 FORWARD
118 124 -245457 -1718200 -52986 -201142 2 0.95 SIL
125 143 -228044 -4332844 -184533 -153653 5 1.05 TEN
144 147 -247742 -990970 -52986 -171520 5 0.81 SIL
148 194 -231355 -10873712 -49359 -159068 3 1.11 METERS
195 502 -176450 -54346772 -49359 -174854 1 0.13 </s>
INFO: search.c(2643): FWDTREE: GO FORWARD TEN METERS (goforward -97490091 (A=-96733294 L=-756797))
INFO: search.c(2555): 1360 words recognized (3/fr)
INFO: search.c(2558): 4724 phones in topsen (9/fr)
INFO: search.c(2560): 3146265 senones evaluated (6255/fr)
INFO: search.c(2562): 29157 channels searched (57/fr), 15390 1st, 9634 last
INFO: search.c(2566): 5146 words for which last channels evaluated (10/fr)
INFO: search.c(2568): 783 candidate words for entering last phone (1/fr)
SFrm EFrm AScr/Frm AScr PathScr BSDiff LatDen PhPerp Word (Bestpath) (goforward)
------------------------------------------------------------------------
63 76 349174 4888443 17438266 -141750 8 1.19 GO
86 117 290375 9292029 26999997 -167232 2 0.89 FORWARD
125 143 282989 5376800 32448937 -153653 5 1.05 TEN
148 194 231355 10873712 43394789 -159068 3 1.11 METERS
INFO: searchlat.c(939): BESTPATH: GO FORWARD TEN METERS (goforward -97741561)
INFO: searchlat.c(942): 17 nodes, 50 links searched
INFO: uttproc.c(507): 5.02 SoS, 3.17 sec elapsed, 0.63 xRT, 3.11 sec CPU, 0.62 xRT
INFO: uttproc.c(550):
INFO: uttproc.c(552): TOTAL Elapsed time 3.17 seconds
INFO: uttproc.c(554): TOTAL CPU time 3.11 seconds
INFO: uttproc.c(556): TOTAL Speech 5.02 seconds
INFO: uttproc.c(559): AVERAGE 0.63 xRT(Elapsed), 0.62 xRT(CPU)
TEST FINISHED
|
Heh turtle. wow. That takes me back... BBC B computers and an archaic plotting program. How proud the teachers were to draw those crappy few lines on the oversized, over expensive paper, if they could see me now...
(they would probably expel me again lol)