Bad performance of pocketsphinx code

[Pablo Caravaca]

Hello! I woud like to post a topic about a problem I am having with an application of pocketsphinx.

Here are the details:

I want to use an input buffer from a microphone to feed the decoder of pocketsphinx. Here is the code i am into:

const char * recognize_from_microphone(cb_t *q){
int16_t buffer[4096];
ps_start_utt(ps);
utt_started = FALSE;

clock_t start, end;
double diff_t;

while(1) {
    int k = cb_get(q, buffer, 4096);

    start = clock();
    ps_process_raw(ps, (const int16_t *)buffer, k, FALSE, FALSE);
    end = clock();

    in_speech = ps_get_in_speech(ps);

    diff_t = ((double) (end - start)) / CLOCKS_PER_SEC;

    printf("Execution time = %f\n", diff_t);
    printf("%d\n",in_speech);

    if (in_speech && !utt_started) {
        utt_started = TRUE;   
        printf("Listening...\n");
    }

    if (!in_speech && utt_started) {
        ps_end_utt(ps); 
        hyp = ps_get_hyp(ps, NULL ); 
        if (hyp != NULL){
            return hyp; 
            break;
        }
    }
    sleep_msec(20);
}

So, the 2 printf functions are saying to me that ps_process_raw is getting around 1 sec for executing each time and in_speech always is returned as ‘1’. I’ve been fighting with the code but still cannot find a solution.

The microphone is working ok, and it is waiting for the cb_get function to be executed to have free space in buffer (producer/consumer solution).

Help is much appreciated. Cheers.

[Nickolay V. Shmyrev]

It is hard to tell, probably your cpu is very slow. Did you try to decode from file first with pocketsphinx_continuous? How long does it take.

[Pablo Caravaca]

Thanks, Nickolay, can you tell me something about this out for continuos.c --inmic yes?

INFO: continous.c(254): Ready....
Execution time = 0.000011
Execution time = 0.000855
Execution time = 0.000814
Execution time = 0.001174
Execution time = 0.000897
Execution time = 0.001875
Execution time = 0.002055
Execution time = 1.281628
INFO: continous.c(273): Listening...
Input overrun, read calls are too rare (non-fatal)
Execution time = 0.000006
Execution time = 0.000004
Execution time = 1.812801
Input overrun, read calls are too rare (non-fatal)
Execution time = 0.000004
Execution time = 0.000003
Execution time = 1.276526
Input overrun, read calls are too rare (non-fatal)
Execution time = 0.000005
Execution time = 0.000005
Execution time = 2.089774
Input overrun, read calls are too rare (non-fatal)
Execution time = 0.000004
Execution time = 0.000003

[Nickolay V. Shmyrev]

I asked you to decode a file and provide the full output as well as cpu info.

[Pablo Caravaca]

Yeah, your right, sorry. I am kindla desperate right now. haha. I'm using Raspberry pi 3 b+, the out for file decoding is:

INFO: ngram_search.c(467): Resized score stack to 200000 entries
INFO: ngram_search.c(459): Resized backpointer table to 10000 entries
INFO: ngram_search_fwdtree.c(949): cand_sf[] increased to 64 entries
INFO: ngram_search.c(467): Resized score stack to 400000 entries
INFO: ngram_search.c(459): Resized backpointer table to 20000 entries
INFO: ngram_search.c(467): Resized score stack to 800000 entries
INFO: ngram_search.c(459): Resized backpointer table to 40000 entries
INFO: ngram_search.c(467): Resized score stack to 1600000 entries
INFO: cmn_live.c(120): Update from < 41.00 -5.29 -0.12 5.09 2.48 -4.07 -1.37 -1.78 -5.08 -2.05 -6.45 -1.42 1.17 >
INFO: cmn_live.c(138): Update to < 37.44 -3.28 3.53 6.57 -8.86 11.49 -12.17 8.89 3.51 -8.78 -1.72 -3.90 -0.12 >
INFO: ngram_search_fwdtree.c(1550): 34051 words recognized (68/fr)
INFO: ngram_search_fwdtree.c(1552): 1868504 senones evaluated (3737/fr)
INFO: ngram_search_fwdtree.c(1556): 11798273 channels searched (23596/fr), 317733 1st, 854816 last
INFO: ngram_search_fwdtree.c(1559): 48426 words for which last channels evaluated (96/fr)
INFO: ngram_search_fwdtree.c(1561): 1032622 candidate words for entering last phone (2065/fr)
INFO: ngram_search_fwdtree.c(1564): fwdtree 30.71 CPU 6.141 xRT
INFO: ngram_search_fwdtree.c(1567): fwdtree 30.83 wall 6.166 xRT
INFO: ngram_search_fwdflat.c(302): Utterance vocabulary contains 845 words
INFO: ngram_search_fwdflat.c(948): 21537 words recognized (43/fr)
INFO: ngram_search_fwdflat.c(950): 947620 senones evaluated (1895/fr)
INFO: ngram_search_fwdflat.c(952): 2242514 channels searched (4485/fr)
INFO: ngram_search_fwdflat.c(954): 117970 words searched (235/fr)
INFO: ngram_search_fwdflat.c(957): 77223 word transitions (154/fr)
INFO: ngram_search_fwdflat.c(960): fwdflat 5.48 CPU 1.095 xRT
INFO: ngram_search_fwdflat.c(963): fwdflat 5.48 wall 1.096 xRT
INFO: ngram_search.c(1197): not found in last frame, using this.498 instead
INFO: ngram_search.c(1250): lattice start node 0 end node this.454
INFO: ngram_search.c(1276): Eliminated 182 nodes before end node
INFO: ngram_search.c(1381): Lattice has 2529 nodes, 41578 links
INFO: ps_lattice.c(1380): Bestpath score: -21097
INFO: ps_lattice.c(1384): Normalizer P(O) = alpha(this:454:498) = -1186561
INFO: ps_lattice.c(1441): Joint P(O,S) = -1363451 P(S|O) = -176890
INFO: ngram_search.c(872): bestpath 1.63 CPU 0.326 xRT
INFO: ngram_search.c(875): bestpath 1.63 wall 0.326 xRT
one yeah yeah and and and yes this
INFO: ngram_search_fwdtree.c(429): TOTAL fwdtree 30.71 CPU 6.153 xRT
INFO: ngram_search_fwdtree.c(432): TOTAL fwdtree 30.83 wall 6.178 xRT
INFO: ngram_search_fwdflat.c(176): TOTAL fwdflat 5.48 CPU 1.098 xRT
INFO: ngram_search_fwdflat.c(179): TOTAL fwdflat 5.48 wall 1.098 xRT
INFO: ngram_search.c(303): TOTAL bestpath 1.63 CPU 0.326 xRT
INFO: ngram_search.c(306): TOTAL bestpath 1.63 wall 0.326 xRT

Obviously I didnt say "one yeah yeah and and and yes this" hahaha. It's 5 seconds file with tv in the back

[Nickolay V. Shmyrev]

The output is incomplete, the model is probably too large. The file format could be also wrong.

[Pablo Caravaca]

This is the rest of the output. It's the standard model as I didn't make any modification.
The format of the file is 16bits, 16Khz little endian

INFO: pocketsphinx.c(152): Parsed model-specific feature parameters from /usr/share/pocketsphinx/model/en-us/en-us/feat.params
Current configuration:
[NAME] [DEFLT] [VALUE]
-agc none none
-agcthresh 2.0 2.000000e+00
-allphone
-allphone_ci yes yes
-alpha 0.97 9.700000e-01
-ascale 20.0 2.000000e+01
-aw 1 1
-backtrace no no
-beam 1e-48 1.000000e-48
-bestpath yes yes
-bestpathlw 9.5 9.500000e+00
-ceplen 13 13
-cmn live batch
-cmninit 40,3,-1 41.00,-5.29,-0.12,5.09,2.48,-4.07,-1.37,-1.78,-5.08,-2.05,-6.45,-1.42,1.17
-compallsen no no
-dict /usr/share/pocketsphinx/model/en-us/cmudict-en-us.dict
-dictcase no no
-dither no no
-doublebw no no
-ds 1 1
-fdict
-feat 1s_c_d_dd 1s_c_d_dd
-featparams
-fillprob 1e-8 1.000000e-08
-frate 100 100
-fsg
-fsgusealtpron yes yes
-fsgusefiller yes yes
-fwdflat yes yes
-fwdflatbeam 1e-64 1.000000e-64
-fwdflatefwid 4 4
-fwdflatlw 8.5 8.500000e+00
-fwdflatsfwin 25 25
-fwdflatwbeam 7e-29 7.000000e-29
-fwdtree yes yes
-hmm /usr/share/pocketsphinx/model/en-us/en-us
-input_endian little little
-jsgf
-keyphrase
-kws
-kws_delay 10 10
-kws_plp 1e-1 1.000000e-01
-kws_threshold 1e-30 1.000000e-30
-latsize 5000 5000
-lda
-ldadim 0 0
-lifter 0 22
-lm /usr/share/pocketsphinx/model/en-us/en-us.lm.bin
-lmctl
-lmname
-logbase 1.0001 1.000100e+00
-logfn
-logspec no no
-lowerf 133.33334 1.300000e+02
-lpbeam 1e-40 1.000000e-40
-lponlybeam 7e-29 7.000000e-29
-lw 6.5 6.500000e+00
-maxhmmpf 30000 30000
-maxwpf -1 -1
-mdef
-mean
-mfclogdir
-min_endfr 0 0
-mixw
-mixwfloor 0.0000001 1.000000e-07
-mllr
-mmap yes yes
-ncep 13 13
-nfft 512 512
-nfilt 40 25
-nwpen 1.0 1.000000e+00
-pbeam 1e-48 1.000000e-48
-pip 1.0 1.000000e+00
-pl_beam 1e-10 1.000000e-10
-pl_pbeam 1e-10 1.000000e-10
-pl_pip 1.0 1.000000e+00
-pl_weight 3.0 3.000000e+00
-pl_window 5 5
-rawlogdir
-remove_dc no no
-remove_noise yes yes
-remove_silence yes yes
-round_filters yes yes
-samprate 16000 1.600000e+04
-seed -1 -1
-sendump
-senlogdir
-senmgau
-silprob 0.005 5.000000e-03
-smoothspec no no
-svspec 0-12/13-25/26-38
-tmat
-tmatfloor 0.0001 1.000000e-04
-topn 4 4
-topn_beam 0 0
-toprule
-transform legacy dct
-unit_area yes yes
-upperf 6855.4976 6.800000e+03
-uw 1.0 1.000000e+00
-vad_postspeech 50 50
-vad_prespeech 20 20
-vad_startspeech 10 10
-vad_threshold 3.0 3.000000e+00
-var
-varfloor 0.0001 1.000000e-04
-varnorm no no
-verbose no no
-warp_params
-warp_type inverse_linear inverse_linear
-wbeam 7e-29 7.000000e-29
-wip 0.65 6.500000e-01
-wlen 0.025625 2.562500e-02

INFO: feat.c(715): Initializing feature stream to type: '1s_c_d_dd', ceplen=13, CMN='batch', VARNORM='no', AGC='none'
INFO: acmod.c(162): Using subvector specification 0-12/13-25/26-38
INFO: mdef.c(518): Reading model definition: /usr/share/pocketsphinx/model/en-us/en-us/mdef
INFO: mdef.c(531): Found byte-order mark BMDF, assuming this is a binary mdef file
INFO: bin_mdef.c(336): Reading binary model definition: /usr/share/pocketsphinx/model/en-us/en-us/mdef
INFO: bin_mdef.c(516): 42 CI-phone, 137053 CD-phone, 3 emitstate/phone, 126 CI-sen, 5126 Sen, 29324 Sen-Seq
INFO: tmat.c(149): Reading HMM transition probability matrices: /usr/share/pocketsphinx/model/en-us/en-us/transition_matrices
INFO: acmod.c(113): Attempting to use PTM computation module
INFO: ms_gauden.c(127): Reading mixture gaussian parameter: /usr/share/pocketsphinx/model/en-us/en-us/means
INFO: ms_gauden.c(242): 42 codebook, 3 feature, size:
INFO: ms_gauden.c(244): 128x13
INFO: ms_gauden.c(244): 128x13
INFO: ms_gauden.c(244): 128x13
INFO: ms_gauden.c(127): Reading mixture gaussian parameter: /usr/share/pocketsphinx/model/en-us/en-us/variances
INFO: ms_gauden.c(242): 42 codebook, 3 feature, size:
INFO: ms_gauden.c(244): 128x13
INFO: ms_gauden.c(244): 128x13
INFO: ms_gauden.c(244): 128x13
INFO: ms_gauden.c(304): 222 variance values floored
INFO: ptm_mgau.c(475): Loading senones from dump file /usr/share/pocketsphinx/model/en-us/en-us/sendump
INFO: ptm_mgau.c(499): BEGIN FILE FORMAT DESCRIPTION
INFO: ptm_mgau.c(562): Rows: 128, Columns: 5126
INFO: ptm_mgau.c(594): Using memory-mapped I/O for senones
INFO: ptm_mgau.c(837): Maximum top-N: 4
INFO: phone_loop_search.c(114): State beam -225 Phone exit beam -225 Insertion penalty 0
INFO: dict.c(320): Allocating 138824 * 20 bytes (2711 KiB) for word entries
INFO: dict.c(333): Reading main dictionary: /usr/share/pocketsphinx/model/en-us/cmudict-en-us.dict
INFO: dict.c(213): Dictionary size 134723, allocated 1016 KiB for strings, 1679 KiB for phones
INFO: dict.c(336): 134723 words read
INFO: dict.c(358): Reading filler dictionary: /usr/share/pocketsphinx/model/en-us/en-us/noisedict
INFO: dict.c(213): Dictionary size 134728, allocated 0 KiB for strings, 0 KiB for phones
INFO: dict.c(361): 5 words read
INFO: dict2pid.c(396): Building PID tables for dictionary
INFO: dict2pid.c(406): Allocating 42^3 * 2 bytes (144 KiB) for word-initial triphones
INFO: dict2pid.c(132): Allocated 21336 bytes (20 KiB) for word-final triphones
INFO: dict2pid.c(196): Allocated 21336 bytes (20 KiB) for single-phone word triphones
INFO: ngram_model_trie.c(354): Trying to read LM in trie binary format
INFO: ngram_search_fwdtree.c(74): Initializing search tree
INFO: ngram_search_fwdtree.c(101): 791 unique initial diphones
INFO: ngram_search_fwdtree.c(186): Creating search channels
INFO: ngram_search_fwdtree.c(323): Max nonroot chan increased to 152609
INFO: ngram_search_fwdtree.c(333): Created 723 root, 152481 non-root channels, 53 single-phone words
INFO: ngram_search_fwdflat.c(157): fwdflat: min_ef_width = 4, max_sf_win = 25
INFO: continous.c(319): ./cross COMPILED ON: Aug 15 2019, AT: 00:38:05

[Pablo Caravaca]

It actually decode fast enough, as you can see in the output from recording from the microphone, isn't it? I am a bit lost here. Appreciate your answers

By reducing the size of buffer (to 1024), I got it working continuously, and by using dict and lm customized with a few words, it seems to be working.... Making progress.... But... Still in_speech is always set as 1....

This is the log file for my last attempt:
INFO: pocketsphinx.c(152): Parsed model-specific feature parameters from /usr/share/pocketsphinx/model/en-us/en-us/feat.params
Current configuration:
[NAME] [DEFLT] [VALUE]
-agc none none
-agcthresh 2.0 2.000000e+00
-allphone
-allphone_ci yes yes
-alpha 0.97 9.700000e-01
-ascale 20.0 2.000000e+01
-aw 1 1
-backtrace no no
-beam 1e-48 1.000000e-48
-bestpath yes yes
-bestpathlw 9.5 9.500000e+00
-ceplen 13 13
-cmn live batch
-cmninit 40,3,-1 41.00,-5.29,-0.12,5.09,2.48,-4.07,-1.37,-1.78,-5.08,-2.05,-6.45,-1.42,1.17
-compallsen no no
-dict /home/pi/modelo_mv/6706.dic
-dictcase no no
-dither no no
-doublebw no no
-ds 1 1
-fdict
-feat 1s_c_d_dd 1s_c_d_dd
-featparams
-fillprob 1e-8 1.000000e-08
-frate 100 100
-fsg
-fsgusealtpron yes yes
-fsgusefiller yes yes
-fwdflat yes yes
-fwdflatbeam 1e-64 1.000000e-64
-fwdflatefwid 4 4
-fwdflatlw 8.5 8.500000e+00
-fwdflatsfwin 25 25
-fwdflatwbeam 7e-29 7.000000e-29
-fwdtree yes yes
-hmm /usr/share/pocketsphinx/model/en-us/en-us
-input_endian little little
-jsgf
-keyphrase
-kws
-kws_delay 10 10
-kws_plp 1e-1 1.000000e-01
-kws_threshold 1e-30 1.000000e-30
-latsize 5000 5000
-lda
-ldadim 0 0
-lifter 0 22
-lm /home/pi/modelo_mv/6706.lm
-lmctl
-lmname
-logbase 1.0001 1.000100e+00
-logfn /home/pi/modelo_mv/log
-logspec no no
-lowerf 133.33334 1.300000e+02
-lpbeam 1e-40 1.000000e-40
-lponlybeam 7e-29 7.000000e-29
-lw 6.5 6.500000e+00
-maxhmmpf 30000 30000
-maxwpf -1 -1
-mdef
-mean
-mfclogdir
-min_endfr 0 0
-mixw
-mixwfloor 0.0000001 1.000000e-07
-mllr
-mmap yes yes
-ncep 13 13
-nfft 512 512
-nfilt 40 25
-nwpen 1.0 1.000000e+00
-pbeam 1e-48 1.000000e-48
-pip 1.0 1.000000e+00
-pl_beam 1e-10 1.000000e-10
-pl_pbeam 1e-10 1.000000e-10
-pl_pip 1.0 1.000000e+00
-pl_weight 3.0 3.000000e+00
-pl_window 5 5
-rawlogdir
-remove_dc no no
-remove_noise yes yes
-remove_silence yes yes
-round_filters yes yes
-samprate 16000 1.600000e+04
-seed -1 -1
-sendump
-senlogdir
-senmgau
-silprob 0.005 5.000000e-03
-smoothspec no no
-svspec 0-12/13-25/26-38
-tmat
-tmatfloor 0.0001 1.000000e-04
-topn 4 4
-topn_beam 0 0
-toprule
-transform legacy dct
-unit_area yes yes
-upperf 6855.4976 6.800000e+03
-uw 1.0 1.000000e+00
-vad_postspeech 50 50
-vad_prespeech 20 20
-vad_startspeech 10 10
-vad_threshold 3.0 3.000000e+00
-var
-varfloor 0.0001 1.000000e-04
-varnorm no no
-verbose no no
-warp_params
-warp_type inverse_linear inverse_linear
-wbeam 7e-29 7.000000e-29
-wip 0.65 6.500000e-01
-wlen 0.025625 2.562500e-02

INFO: feat.c(715): Initializing feature stream to type: '1s_c_d_dd', ceplen=13, CMN='batch', VARNORM='no', AGC='none'
INFO: acmod.c(162): Using subvector specification 0-12/13-25/26-38
INFO: mdef.c(518): Reading model definition: /usr/share/pocketsphinx/model/en-us/en-us/mdef
INFO: mdef.c(531): Found byte-order mark BMDF, assuming this is a binary mdef file
INFO: bin_mdef.c(336): Reading binary model definition: /usr/share/pocketsphinx/model/en-us/en-us/mdef
INFO: bin_mdef.c(516): 42 CI-phone, 137053 CD-phone, 3 emitstate/phone, 126 CI-sen, 5126 Sen, 29324 Sen-Seq
INFO: tmat.c(149): Reading HMM transition probability matrices: /usr/share/pocketsphinx/model/en-us/en-us/transition_matrices
INFO: acmod.c(113): Attempting to use PTM computation module
INFO: ms_gauden.c(127): Reading mixture gaussian parameter: /usr/share/pocketsphinx/model/en-us/en-us/means
INFO: ms_gauden.c(242): 42 codebook, 3 feature, size:
INFO: ms_gauden.c(244): 128x13
INFO: ms_gauden.c(244): 128x13
INFO: ms_gauden.c(244): 128x13
INFO: ms_gauden.c(127): Reading mixture gaussian parameter: /usr/share/pocketsphinx/model/en-us/en-us/variances
INFO: ms_gauden.c(242): 42 codebook, 3 feature, size:
INFO: ms_gauden.c(244): 128x13
INFO: ms_gauden.c(244): 128x13
INFO: ms_gauden.c(244): 128x13
INFO: ms_gauden.c(304): 222 variance values floored
INFO: ptm_mgau.c(475): Loading senones from dump file /usr/share/pocketsphinx/model/en-us/en-us/sendump
INFO: ptm_mgau.c(499): BEGIN FILE FORMAT DESCRIPTION
INFO: ptm_mgau.c(562): Rows: 128, Columns: 5126
INFO: ptm_mgau.c(594): Using memory-mapped I/O for senones
INFO: ptm_mgau.c(837): Maximum top-N: 4
INFO: phone_loop_search.c(114): State beam -225 Phone exit beam -225 Insertion penalty 0
INFO: dict.c(320): Allocating 4123 * 20 bytes (80 KiB) for word entries
INFO: dict.c(333): Reading main dictionary: /home/pi/modelo_mv/6706.dic
INFO: dict.c(213): Dictionary size 22, allocated 0 KiB for strings, 0 KiB for phones
INFO: dict.c(336): 22 words read
INFO: dict.c(358): Reading filler dictionary: /usr/share/pocketsphinx/model/en-us/en-us/noisedict
INFO: dict.c(213): Dictionary size 27, allocated 0 KiB for strings, 0 KiB for phones
INFO: dict.c(361): 5 words read
INFO: dict2pid.c(396): Building PID tables for dictionary
INFO: dict2pid.c(406): Allocating 42^3 * 2 bytes (144 KiB) for word-initial triphones
INFO: dict2pid.c(132): Allocated 21336 bytes (20 KiB) for word-final triphones
INFO: dict2pid.c(196): Allocated 21336 bytes (20 KiB) for single-phone word triphones
INFO: ngram_model_trie.c(354): Trying to read LM in trie binary format
INFO: ngram_model_trie.c(365): Header doesn't match
INFO: ngram_model_trie.c(177): Trying to read LM in arpa format
INFO: ngram_model_trie.c(193): LM of order 3
INFO: ngram_model_trie.c(195): #1-grams: 21
INFO: ngram_model_trie.c(195): #2-grams: 39
INFO: ngram_model_trie.c(195): #3-grams: 39
INFO: lm_trie.c(474): Training quantizer
INFO: lm_trie.c(482): Building LM trie
INFO: ngram_search_fwdtree.c(74): Initializing search tree
INFO: ngram_search_fwdtree.c(101): 20 unique initial diphones
INFO: ngram_search_fwdtree.c(186): Creating search channels
INFO: ngram_search_fwdtree.c(323): Max nonroot chan increased to 175
INFO: ngram_search_fwdtree.c(333): Created 20 root, 47 non-root channels, 5 single-phone words
INFO: ngram_search_fwdflat.c(157): fwdflat: min_ef_width = 4, max_sf_win = 25
INFO: cmn_live.c(88): Update from < 41.00 -5.29 -0.12 5.09 2.48 -4.07 -1.37 -1.78 -5.08 -2.05 -6.45 -1.42 1.17 >
INFO: cmn_live.c(105): Update to < 67.99 -11.31 -6.89 -6.54 -0.91 -0.01 -1.91 -6.54 -0.47 -5.15 -4.71 -2.68 0.13 >
INFO: cmn_live.c(88): Update from < 67.99 -11.31 -6.89 -6.54 -0.91 -0.01 -1.91 -6.54 -0.47 -5.15 -4.71 -2.68 0.13 >
INFO: cmn_live.c(105): Update to < 68.23 -11.15 -6.47 -6.57 -0.67 -0.20 -1.93 -6.66 -0.56 -4.97 -4.78 -2.83 -0.09 >

[Pablo Caravaca]

Welll, I got the continue.c working with a smaller model, and a smaller buffer.
The buffer is receiving ok and functions are executing on time and the audio input is surely getting sampled at 16 Khz, 16bit, little endian. However, I am not recognizing any word (in_speech always 0), what is strange, as the code is practically the same.

I'm doubting about config struct, but seems ok to me....

Here is the code:

#include <stdio.h>
#include <string.h>
#include <pocketsphinx/pocketsphinx.h>
#include <sphinxbase/ad.h>
#include <sphinxbase/err.h>
#include "CBuffer/cbuffer.h"
#include <time.h>

extern "C" cb_t q;

static const arg_t cont_args_def[] = {
    POCKETSPHINX_OPTIONS,
    /* Argument file. */
    {"-argfile", ARG_STRING, NULL, "Argument file giving extra arguments."},
    {"-adcdev", ARG_STRING, NULL, "Name of audio device to use for input."},
    {"-infile", ARG_STRING, NULL, "Audio file to transcribe."},
    {"-inmic", ARG_BOOLEAN, "no", "Transcribe audio from microphone."},
    {"-time", ARG_BOOLEAN, "no", "Print word times in file transcription."},
    CMDLN_EMPTY_OPTION};

const char * recognize_from_microphone();

ps_decoder_t *ps;       //  Decoder structure
cmd_ln_t *config;       //  Configuration for decoder

uint8 utt_started;      //  Flags for utterance started and speech is producing
int32 k;                           //   Number of frames in audio buffer
char const *hyp;                   //   Hypothesis for given speech
char const *decoded_speech;

void *t_sphinx (void *arg) {

    config = cmd_ln_init(NULL,
            cont_args_def, TRUE,
            "-lm", "/home/pi/modelo_mv/6706.lm",
            "-dict", "/home/pi/modelo_mv/6706.dic",
            "-kws_threshold", "1e-20",
            "-keyphrase", "MATRIX",
            NULL);

    ps_default_search_args(config);
    ps = ps_init(config);                                                        // initialize the pocketsphinx decoder

    while(1){
        decoded_speech = recognize_from_microphone();                 // call the function to capture and decode speech
        printf("You Said: %s\n", decoded_speech);                               // send decoded speech to screen
    }

    ps_free(ps);
}

static void sleep_msec(int32 ms) {
    struct timeval tmo;
    tmo.tv_sec = 0;
    tmo.tv_usec = ms * 1000;

    select(0, NULL, NULL, NULL, &tmo);
}

const char * recognize_from_microphone(){
    int16_t buffer[2046];
    if (ps_start_utt(ps) < 0) E_FATAL("Failed to start utterance\n");
    utt_started = FALSE;
    printf("Ready...\n");

    clock_t start, end;
    double diff_t;
    int k,j;

    while(1) {
        k = cb_get(&q, buffer, 2046);   // capture the number of frames in the audio buffer

        start = clock();
        ps_process_raw(ps, (const int16_t *)buffer, k, FALSE, FALSE);  // send the audio buffer to the pocketsphinx decoder
        const int8_t in_speech = ps_get_in_speech(ps);            // test to see if speech is being detected
        end = clock();

        if (j ==40){
            diff_t = ((double) (end - start)) / CLOCKS_PER_SEC;
            printf("%d\t",in_speech);
            printf("%f\n",diff_t);
            j=0;
        }
        else
            j++;

        if (in_speech && !utt_started) {             // if speech has started and utt_started flag is false
            utt_started = TRUE;                      // then set the flag
            printf("Listening...\n");
        }

        if (!in_speech && utt_started) {             // if speech has ended and the utt_started flag is true
            ps_end_utt(ps);                          // then mark the end of the utterance
            hyp = ps_get_hyp(ps, NULL );             // query pocketsphinx for "hypothesis" of decoded statement
            return hyp;                              // the function returns the hypothesis
            break;                               // exit the while loop and return to main

        }
        sleep_msec(10);
    }
}