Does pocketsphinx LiveSpeech lose accuracy over the time?

[Paulo Ferreira]

I am building a program with pocketsphinx and python this is the code i have so far:

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

#!/usr/bin/env python import os, time from socket import * from pocketsphinx import LiveSpeech, get_model_path model_path = get_model_path() socket = socket(AF_INET, SOCK_STREAM) #cria um socket host = 'HASTurtle' port = 8421 vel_linear = 0.25 vel_angular = 1.5 speech = LiveSpeech( verbose=False, sampling_rate=16000, buffer_size=2048, no_search=False, full_utt=False, lm=False, hmm= os.path.join(model_path,'en-us'), #lm= '/home/paulo/PycharmProjects/VoiceRecognition/Data/TAR4858/4858.lm', dict = #'/home/vitor/Documentos/Univ/3_Ano/2_Semestre/Projeto/Projeto/VoiceRecognition/Data/Robot/Robot.dic', '/home/paulo/PycharmProjects/VoiceRecognition/Data/Robot/Robot.dic', kws = #'/home/vitor/Documentos/Univ/3_Ano/2_Semestre/Projeto/Projeto/VoiceRecognition/Data/Robot/Robot_Keyphrase.list' '/home/paulo/PycharmProjects/VoiceRecognition/Data/Robot/Robot_Keyphrase.list' ) def parseVoice(frase): if frase == 'MOVE FORWARD ': valores = ' '.join([str(vel_linear),str(0)]) socket.sendall(''.join([valores,'\n']).encode()) elif frase == 'ROTATE LEFT ': valores = ' '.join([str(0),str(-vel_angular)]) socket.sendall(''.join([valores, '\n']).encode()) elif frase == 'ROTATE RIGHT ': valores = ' '.join([str(0),str(vel_angular)]) socket.sendall(''.join([valores, '\n']).encode()) elif frase == 'STOP MOVING ': socket.sendall('0 0\n'.encode()) elif frase == 'STOP LISTENING ': return -1 return 1 socket.connect(('localhost',port)) print 'A enviar talker ...' time.sleep(1) socket.send('talker\n'.decode()) print 'Enviou talker' print 'Listening...' for phrase in speech: frase = str(phrase) if parseVoice(frase) == -1: socket.close() break data = socket.recv(1024).decode() print data, print 'Listening...'

Keyphrase list:

ROTATE RIGHT /1e-40/
ROTATE LEFT /1e-35/
MOVE FORWARD /1e-50/
STOP MOVING /1e-10/
STOP LISTENING /1e-20/

When i first run this code and speak everything works perfect but when i give more than 5 or 6 commands i need to say the command many times until it gets recognized and sometimes it doesn't recognize at all. Should LiveSpeech be reseted or something like that?

[Nickolay V. Shmyrev]

Add verbose = True in arguments and track how CMN values change with time. You need to make sure that voice activity detection works and it only reacts to voice, not to the noise. If it reacts to the noise you need to raise the voice activity threshold.