Pocketsphinx Talk on Pycon

Talk by our David

4 Responses to “Pocketsphinx Talk on Pycon”

  1. Antonello Mangone says:

    Hi, I followed your talk and I tired myself to do it, but I have always the same problem ….
    …..
    INFO: cmn.c(175): CMN: 49.22 3.71 0.49 1.63 1.12 0.01 0.13 0.06 -0.38 -0.39 -0.50 -0.23 0.19
    INFO: ngram_search.c(376): Resized score stack to 200000 entries
    INFO: ngram_search.c(368): Resized backpointer table to 10000 entries
    INFO: ngram_search.c(376): Resized score stack to 400000 entries
    INFO: ngram_search.c(368): Resized backpointer table to 20000 entries
    INFO: ngram_search_fwdtree.c(1450): 12857 words recognized (18/fr)
    INFO: ngram_search_fwdtree.c(1452): 2016977 senones evaluated (2857/fr)
    INFO: ngram_search_fwdtree.c(1454): 1859926 channels searched (2634/fr), 268897 1st, 386343 last
    INFO: ngram_search_fwdtree.c(1458): 25626 words for which last channels evaluated (36/fr)
    INFO: ngram_search_fwdtree.c(1461): 117127 candidate words for entering last phone (165/fr)
    INFO: ngram_search_fwdflat.c(840): 10012 words recognized (14/fr)
    INFO: ngram_search_fwdflat.c(842): 735735 senones evaluated (1042/fr)
    INFO: ngram_search_fwdflat.c(844): 754140 channels searched (1068/fr)
    INFO: ngram_search_fwdflat.c(846): 47699 words searched (67/fr)
    INFO: ngram_search_fwdflat.c(848): 25970 word transitions (36/fr)
    WARNING: “ngram_search.c”, line 1000: not found in last frame, using HUTTON instead
    INFO: ngram_search.c(1046): lattice start node .0 end node HUTTON.448
    INFO: ps_lattice.c(1225): Normalizer P(O) = alpha(HUTTON:448:704) = -7061076
    INFO: ps_lattice.c(1263): Joint P(O,S) = -7070637 P(S|O) = -9561
    (’THE ERA IN NEW ,COMMA THE TURNAROUND AT HUTTON’, ‘000000000′, -136948675)

    …. the file i wrote is ….

    #!/usr/bin/env python
    # -*- codiing:utf-8 -*-

    import pocketsphinx as ps
    import sphinxbase

    decoder = ps.Decoder(hmm=”/usr/share/pocketsphinx/model/hmm/wsj1″,lm=”/usr/share/pocketsphinx/model/lm/wsj/wlist5o.3e-7.vp.tg.lm.DMP”,dict=”/usr/share/pocketsphinx/model/lm/wsj/wlist5o.dic”)

    fh = file(”example.wav”,”rb”)
    fh.seek(0)
    decoder.decode_raw(fh)
    print decoder.get_hyp()

    ……………………………….

    Can you help me please ?
    Best Regards

  2. admin says:

    Antonello, I don’t see any problem here, can you elaborate?

    If you are about wrong result, probably your audio file has wrong sample rate. It must be 16khz 16bit mono.

    P.S. It’s better ot move this discussion to sourceforge forum where uspport takes place.

  3. Umaid says:

    Does your speech recognition has support on iphone. As i m interested in it in using in some iphone applcation development

  4. admin says:

    Don’t you see the post below? It’s named “pocketspinx for IPhone”