Problem with PocketSphinx using continuous AM

[mmarge]

Hi,

I'm getting the following error when trying to run a (successfully installed)
pocketsphinx_batch with a continuous acoustic model:
pocketsphinx_batch: ngram_search.c:707: ngram_compute_seg_score: Assertion
`start_score > ((int)0xE0000000)' failed.
Abort

Do you know what could be going on?

I'm running PocketSphinx 0.7 on Fedora 10.

Here's the relevant set of parameters I used for PocketSphinx:

-samprate 16000
-cmn current
-agc none
-logfn ./desktop.log

Here are the parameters in feat.params for the continuous acoustic model:
-alpha 0.97
-dither yes
-doublebw no
-nfilt 40
-ncep 13
-lowerf 200
-upperf 6700
-nfft 512
-wlen 0.0256
-transform dct
-feat 1s_c_d_dd
-agc none
-cmn current
-varnorm no

Here's the output of the log file:

INFO: cmd_ln.c(559): Parsing command line:
\
-alpha 0.97 \
-dither yes \
-doublebw no \
-nfilt 40 \
-ncep 13 \
-lowerf 200 \
-upperf 6700 \
-nfft 512 \
-wlen 0.0256 \
-transform dct \
-feat 1s_c_d_dd \
-agc none \
-cmn current \
-varnorm no

Current configuration:

-agc none none
-agcthresh 2.0 2.000000e+00
-alpha 0.97 9.700000e-01
-ceplen 13 13
-cmn current current
-cmninit 8.0 8.0
-dither no yes
-doublebw no no
-feat 1s_c_d_dd 1s_c_d_dd
-frate 100 100
-input_endian little little
-lda
-ldadim 0 0
-lifter 0 0
-logspec no no
-lowerf 133.33334 2.000000e+02
-ncep 13 13
-nfft 512 512
-nfilt 40 40
-remove_dc no no
-round_filters yes yes
-samprate 16000 1.600000e+04
-seed -1 -1
-smoothspec no no
-svspec
-transform legacy dct
-unit_area yes yes
-upperf 6855.4976 6.700000e+03
-varnorm no no
-verbose no no
-warp_params
-warp_type inverse_linear inverse_linear
-wlen 0.025625 2.560000e-02

INFO: acmod.c(242): Parsed model-specific feature parameters from
/usr1/mrmarge/thesiswork/asr/testdata/200_6700.cd_cont_4000/feat.params
INFO: fe_interface.c(289): You are using the internal mechanism to generate
the seed.
INFO: feat.c(697): Initializing feature stream to type: '1s_c_d_dd',
ceplen=13, CMN='current', VARNORM='no', AGC='none'
INFO: cmn.c(142): mean= 12.00, mean= 0.0
INFO: mdef.c(520): Reading model definition:
/usr1/mrmarge/thesiswork/asr/testdata/200_6700.cd_cont_4000/mdef
INFO: bin_mdef.c(173): Allocating 70048 * 8 bytes (547 KiB) for CD tree
INFO: tmat.c(205): Reading HMM transition probability matrices: /usr1/mrmarge/
thesiswork/asr/testdata/200_6700.cd_cont_4000/transition_matrices
INFO: acmod.c(117): Attempting to use SCHMM computation module
INFO: ms_gauden.c(198): Reading mixture gaussian parameter:
/usr1/mrmarge/thesiswork/asr/testdata/200_6700.cd_cont_4000/means
INFO: ms_gauden.c(292): 4132 codebook, 1 feature, size:
INFO: ms_gauden.c(294): 32x39
INFO: ms_gauden.c(198): Reading mixture gaussian parameter:
/usr1/mrmarge/thesiswork/asr/testdata/200_6700.cd_cont_4000/variances
INFO: ms_gauden.c(292): 4132 codebook, 1 feature, size:
INFO: ms_gauden.c(294): 32x39
INFO: ms_gauden.c(354): 52 variance values floored
INFO: acmod.c(119): Attempting to use PTHMM computation module
INFO: ms_gauden.c(198): Reading mixture gaussian parameter:
/usr1/mrmarge/thesiswork/asr/testdata/200_6700.cd_cont_4000/means
INFO: ms_gauden.c(292): 4132 codebook, 1 feature, size:
INFO: ms_gauden.c(294): 32x39
INFO: ms_gauden.c(198): Reading mixture gaussian parameter:
/usr1/mrmarge/thesiswork/asr/testdata/200_6700.cd_cont_4000/variances
INFO: ms_gauden.c(292): 4132 codebook, 1 feature, size:
INFO: ms_gauden.c(294): 32x39
INFO: ms_gauden.c(354): 52 variance values floored
INFO: ptm_mgau.c(800): Number of codebooks exceeds 256: 4132
INFO: acmod.c(121): Falling back to general multi-stream GMM computation
INFO: ms_gauden.c(198): Reading mixture gaussian parameter:
/usr1/mrmarge/thesiswork/asr/testdata/200_6700.cd_cont_4000/means
INFO: ms_gauden.c(292): 4132 codebook, 1 feature, size:
INFO: ms_gauden.c(294): 32x39
INFO: ms_gauden.c(198): Reading mixture gaussian parameter:
/usr1/mrmarge/thesiswork/asr/testdata/200_6700.cd_cont_4000/variances
INFO: ms_gauden.c(292): 4132 codebook, 1 feature, size:
INFO: ms_gauden.c(294): 32x39
INFO: ms_gauden.c(354): 52 variance values floored
INFO: ms_senone.c(160): Reading senone mixture weights:
/usr1/mrmarge/thesiswork/asr/testdata/200_6700.cd_cont_4000/mixture_weights
INFO: ms_senone.c(211): Truncating senone logs3(pdf) values by 10 bits
INFO: ms_senone.c(218): Not transposing mixture weights in memory
INFO: ms_senone.c(277): Read mixture weights for 4132 senones: 1 features x 32
codewords
INFO: ms_senone.c(331): Mapping senones to individual codebooks
INFO: ms_mgau.c(122): The value of topn: 4
INFO: dict.c(306): Allocating 10371 * 32 bytes (324 KiB) for word entries
INFO: dict.c(321): Reading main dictionary:
/net/ginkgo/usr0/lqin/wsj_si284/etc/tcb05cnp.dic
INFO: dict.c(212): Allocated 47 KiB for strings, 74 KiB for phones
INFO: dict.c(324): 6232 words read
INFO: dict.c(330): Reading filler dictionary:
/usr1/mrmarge/thesiswork/asr/testdata/200_6700.cd_cont_4000/noisedict
INFO: dict.c(212): Allocated 0 KiB for strings, 0 KiB for phones
INFO: dict.c(333): 43 words read
INFO: dict2pid.c(396): Building PID tables for dictionary
INFO: dict2pid.c(404): Allocating 44^3 * 2 bytes (166 KiB) for word-initial
triphones
INFO: dict2pid.c(131): Allocated 46816 bytes (45 KiB) for word-final triphones
INFO: dict2pid.c(195): Allocated 46816 bytes (45 KiB) for single-phone word
triphones
INFO: ngram_model_arpa.c(77): No \data\ mark in LM file
INFO: ngram_model_dmp.c(142): Will use memory-mapped I/O for LM file
INFO: ngram_model_dmp.c(196): ngrams 1=4989, 2=1639687, 3=2684151
INFO: ngram_model_dmp.c(242): 4989 = LM.unigrams(+trailer) read
WARNING: "ngram_model_dmp.c", line 253: -mmap specified, but tseg_base is not
word-aligned. Will not memory-map.
INFO: ngram_model_dmp.c(291): 1639687 = LM.bigrams(+trailer) read
INFO: ngram_model_dmp.c(317): 2684151 = LM.trigrams read
INFO: ngram_model_dmp.c(342): 25453 = LM.prob2 entries read
INFO: ngram_model_dmp.c(362): 14719 = LM.bo_wt2 entries read
INFO: ngram_model_dmp.c(382): 26030 = LM.prob3 entries read
INFO: ngram_model_dmp.c(410): 3203 = LM.tseg_base entries read
INFO: ngram_model_dmp.c(466): 4989 = ascii word strings read
INFO: ngram_search_fwdtree.c(99): 440 unique initial diphones
WARNING: "ngram_search_fwdtree.c", line 111: Filler word 6273 = ++UM++ has
more than one phone, ignoring it.
INFO: ngram_search_fwdtree.c(147): 0 root, 0 non-root channels, 53 single-
phone words
INFO: ngram_search_fwdtree.c(186): Creating search tree
INFO: ngram_search_fwdtree.c(191): before: 0 root, 0 non-root channels, 53
single-phone words
INFO: ngram_search_fwdtree.c(326): after: max nonroot chan increased to 14037
INFO: ngram_search_fwdtree.c(338): after: 440 root, 13909 non-root channels,
52 single-phone words
INFO: ngram_search_fwdflat.c(156): fwdflat: min_ef_width = 4, max_sf_win = 25
INFO: cmn.c(175): CMN: 6.84 0.13 -0.08 0.22 -0.03 -0.10 -0.22 -0.28 -0.14
-0.08 -0.20 -0.12 -0.15
INFO: ngram_search.c(466): Resized backpointer table to 10000 entries
INFO: ngram_search.c(466): Resized backpointer table to 20000 entries
INFO: ngram_search_fwdtree.c(1549): 14385 words recognized (21/fr)
INFO: ngram_search_fwdtree.c(1551): 1062811 senones evaluated (1527/fr)
INFO: ngram_search_fwdtree.c(1553): 994948 channels searched (1429/fr), 242177
1st, 95260 last
INFO: ngram_search_fwdtree.c(1557): 26005 words for which last channels
evaluated (37/fr)
INFO: ngram_search_fwdtree.c(1560): 48649 candidate words for entering last
phone (69/fr)
INFO: ngram_search_fwdtree.c(1562): fwdtree 4.37 CPU 0.628 xRT
INFO: ngram_search_fwdtree.c(1565): fwdtree 4.37 wall 0.628 xRT
INFO: ngram_search_fwdflat.c(305): Utterance vocabulary contains 145 words
INFO: ngram_search_fwdflat.c(940): 13902 words recognized (20/fr)
INFO: ngram_search_fwdflat.c(942): 103646 senones evaluated (149/fr)
INFO: ngram_search_fwdflat.c(944): 100589 channels searched (144/fr)
INFO: ngram_search_fwdflat.c(946): 30637 words searched (44/fr)
INFO: ngram_search_fwdflat.c(948): 8413 word transitions (12/fr)
INFO: ngram_search_fwdflat.c(951): fwdflat 0.35 CPU 0.050 xRT
INFO: ngram_search_fwdflat.c(954): fwdflat 0.35 wall 0.050 xRT
INFO: ngram_search.c(1201): not found in last frame, using <sil>.694
instead
INFO: ngram_search.c(1253): lattice start node .0 end node <sil>.516
INFO: ngram_search.c(1281): Eliminated 3030 nodes before end node
INFO: ngram_search.c(1386): Lattice has 4677 nodes, 30154 links
INFO: ps_lattice.c(1352): Normalizer P(O) = alpha(<sil>:516:694) = -1595596
INFO: ps_lattice.c(1390): Joint P(O,S) = -1612563 P(S|O) = -16967
INFO: ngram_search.c(875): bestpath 0.14 CPU 0.020 xRT
INFO: ngram_search.c(878): bestpath 0.14 wall 0.020 xRT
INFO: batch.c(760): wsj-5k-feats/Matt_Kinect_1m_200_6700/0_kinect: 6.95
seconds speech, 4.86 seconds CPU, 4.86 seconds wall
INFO: batch.c(762): wsj-5k-feats/Matt_Kinect_1m_200_6700/0_kinect: 0.70 xRT
(CPU), 0.70 xRT (elapsed)
INFO: cmn.c(175): CMN: 6.63 0.12 -0.05 0.13 -0.03 -0.05 -0.28 -0.27 -0.14
-0.08 -0.16 -0.10 -0.15
INFO: ngram_search_fwdtree.c(1549): 12206 words recognized (23/fr)
INFO: ngram_search_fwdtree.c(1551): 856007 senones evaluated (1630/fr)
INFO: ngram_search_fwdtree.c(1553): 806907 channels searched (1536/fr), 190707
1st, 61226 last
INFO: ngram_search_fwdtree.c(1557): 21307 words for which last channels
evaluated (40/fr)
INFO: ngram_search_fwdtree.c(1560): 41912 candidate words for entering last
phone (79/fr)
INFO: ngram_search_fwdtree.c(1562): fwdtree 3.59 CPU 0.684 xRT
INFO: ngram_search_fwdtree.c(1565): fwdtree 3.59 wall 0.684 xRT
INFO: ngram_search_fwdflat.c(305): Utterance vocabulary contains 68 words
INFO: ngram_search_fwdflat.c(940): 11700 words recognized (22/fr)
INFO: ngram_search_fwdflat.c(942): 69932 senones evaluated (133/fr)
INFO: ngram_search_fwdflat.c(944): 57811 channels searched (110/fr)
INFO: ngram_search_fwdflat.c(946): 20648 words searched (39/fr)
INFO: ngram_search_fwdflat.c(948): 3988 word transitions (7/fr)
INFO: ngram_search_fwdflat.c(951): fwdflat 0.21 CPU 0.039 xRT
INFO: ngram_search_fwdflat.c(954): fwdflat 0.21 wall 0.039 xRT
INFO: ngram_search.c(1201): </sil></sil> not found in last frame, using ++BEEP++.523
instead
INFO: ngram_search.c(1253): lattice start node .0 end node ++BEEP++.383
INFO: ngram_search.c(1281): Eliminated 1013 nodes before end node </sil>

I get only one hypothesis, when I should have received 25 hypotheses.

Thanks for your help!
Matt

[Nickolay V. Shmyrev]

Sorry, it's not quite clear where the assertion is failed, it is missed in the
log. Can you please share the full log?

To make it easier to diagnose your problem and to get help you can always
share the data which will enable to reproduce your problem.

[mmarge]

Thanks for replying so quickly - I am pretty sure I pasted the full log - that
was the entirety of contents in desktop.log (this is the file which saves the
log as specified in PocketSphinx's command line params.

Do you have any ideas what could be the problem? Thanks!

[Nickolay V. Shmyrev]

Do you have any ideas what could be the problem? Thanks!

No, without data it's hard to say anything.