Loading another HMM model

[Tania Mendonca]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74

#!/usr/bin/python from os import environ, path from glob import glob import os from pocketsphinx.pocketsphinx import * from sphinxbase.sphinxbase import * MODELDIR = "/home/tania/trial/sylabble" DATADIR = "/home/tania/trial/sylabble/wav/kannadatest_clstk" # Create a decoder with certain model config = Decoder.default_config() config.set_string('-hmm', path.join(MODELDIR, 'model_parameters/sylabble.cd_cont_200')) config.set_string('-lm', path.join(MODELDIR, 'etc/sylabble.lm.DMP')) config.set_string('-dict', path.join(MODELDIR, 'etc/sylabble.dic')) config.set_float('-beam', 1e-80) config.set_float('-fwdflatbeam', 1e-80) config.set_float('-fwdflatlw', 10) config.set_float('-fwdflatwbeam',1e-40) config.set_float('-lpbeam',1e-80) config.set_float('-lponlybeam',1e-80) config.set_float('-lw',10) config.set_float('-pbeam',1e-80) config.set_float('-wbeam',1e-40) config.set_float('-wip',2e-01) #config.set_string('-logfn', "/dev/null") i=0 # Decode streaming data. decoder = Decoder(config) while i<2: decoder.start_utt() stream = open(path.join(DATADIR,'kan_0702.wav' ), 'rb') while True: buf = stream.read() if buf: decoder.process_raw(buf, False, True) else: break i=i+1 decoder.end_utt() hypothesis = decoder.hyp() s=hypothesis.hypstr #print(s) z="".join(s.split()) z=z.replace("$^"," ") z=z.replace("^"," ") z=z.replace("$"," ") z=z.replace("_","") z=z.strip() print (z) i=0 DATADIR1="/home/tania/z1/kannada1-2" lm = NGramModel(config, decoder.get_logmath(), path.join(DATADIR1, 'etc/kannada1-2.lm.DMP')) decoder.set_lm('kannada1-2', lm) decoder.set_search('kannada1-2') decoder.load_dict("/home/tania/z1/kannada1-2/etc/kannada1-2.dic",None,None) while i<2: decoder.start_utt() stream = open(path.join(DATADIR, 'kan_0702.wav'), 'rb') while True: buf = stream.read() if buf: decoder.process_raw(buf, False, True) else: break i=i+1 decoder.end_utt() hypothesis = decoder.hyp() s=hypothesis.hypstr print (s)

In the above code i was able to set a new LM and load a new dict. How can i update the hmm model in the code from sylabble.cd_cont_200 to kannada1-2.cd_cont_200 without creating a new decoder object? Is it posibble?
Without changing the HMM model its still able to give proper results I'm not sure how
and the while loop i used the loop twice because in the first decoding i dont get the right result. i tried changing the changing the channel parameters too but i dont the right decoded sentence in first shot. Please help me out

[Nickolay V. Shmyrev]

Not possible.

[Tania Mendonca]

Any idea as to how its getting the right results for word decoding from the sylabble based HMM? Is it because the HMM is phone based though i have made a sylabble based?