Artigos de Abratel Telecom & InfoReconhecimento de voz com asterisk2012-01-21 19:01:28 Ulisses Féres

A idéia é utilizar EAGI para controle do canal de entrada de áudio em conjunto como File Descriptor, o Asterisk entrega o áudio em formato RAW diretamente no FileDescriptor 3, então podemos utilizar esta informação da maneira que acharmosconveniente, para este caso a manipulação se torna muito prática, o que medesprende totalmente das APP’s prontas para gravações inseridas no Asterisk Ex.Record, nada melhor do que ser livre para voar, é claro várias análises se tornampossíveis com isso e o leque de aplicações possíveis se tornam infinitas.

Estou usando novamente o módulo audiolab para efetuar o encode do áudio emFLAC, caso exista alguma dificuldade para a instalação deste módulo podereipensar em adaptar o código para uso externo do sox ou flac.

Como ele funciona?

Atende uma ligaçãoO usuário tem no máximo 10 segundos para efetuar a falaCaso nao encontre atividade de voz encerra com timeoutEstratégia para atividade de voz verdadeira para os seguintes valores RMS > 15 ePitch > 75Se atividade for encontrada o usuário poderá falar por no máximo 10 segundosO script verifica blocos em tempo real com amostras de 1 em 1 segundo e verificase a fala cessouCaso sim o script interrompe a gravação automáticamente e envia o que foigravado para o googleCaso não o script continua o seu curso até seu máximo de 10 segundosApos encontrada a resposta da fala no google o script seta a variável“GoogleUtterance”

Instalacao:

Dependencies:

apt-get install python-matplotlibapt-get install python-numpyapt-get install python-scipyapt-get install python-dev python-setuptools libsndfile-dev

Download and install audiolab from:

http://pypi.python.org/pypi/scikits.audiolab/

Example how use in dialplan from Asterisk:Extensions.conf

exten=>_11111111,1,Answer()exten=>_11111111,n,eagi,pahh.pyexten=>_11111111,n,GotoIf($[${EXISTS(${GoogleUtterance})}]?hello:bye)exten=>_11111111,n(hello),NoOP(You Said = ${GoogleUtterance})exten=>_11111111,n(bye),Hangup()

Fiz um reconhecimento com comparacao:exten=>_1,1,Answer()exten=>_1,n,eagi(pahh.py)exten=>_1,n,GotoIf($[${EXISTS(${GoogleUtterance})}]?hello:bye)exten=>_1,n(hello),NoOP(You Said = ${GoogleUtterance})exten=>_1,n(hello),GotoIf($["${GoogleUtterance}" = "9 0 8"]?acertei,s,1)exten=>_1,n(hello),GotoIf($["${GoogleUtterance}" = "9 0 5"]?acertei,s,100)exten=>_1,n(hello),GotoIf($["${GoogleUtterance}" = "9 1 3"]?acertei,s,200)exten=>_1,n(bye),Hangup()

; tratei a comparacao:[acertei]exten => s,1,Dial(DAHDI/8,20)exten => s,100,Dial(DAHDI/5,20)exten => s,200,Dial(DAHDI/13,20)

Criar o script com nome pahh.py e colocar na pasta /var/lib/asterisk/agi-binEfetuar o comando chmod +x /var/lib/asterisk/agi-bin/pahh.py

Script pahh.py abaixo:#!/usr/bin/python#Copyright (c) 2012, Eng Eder de Souza#Accessing the Google API for speech recognition With Asterisk!#Eng Eder de Souza#date 15/01/2012#http://ederwander.wordpress.com/2012/01/16/google-speech-python-asterisk/## This program is free software, distributed under the terms of# the GNU General Public License Version 2. See the COPYING file# at the top of the source tree.##Revision 0.2#History:#18/01/2012 bug fix in local variable declaration#19/01/2012 suport for old python interpretator#19/01/2012 removed matplotlib dependencies#19/01/2012 Submission of warnings DeprecationWarning and UserWarning

import warningswarnings.simplefilter("ignore", DeprecationWarning)warnings.simplefilter("ignore", UserWarning)from scikits.audiolab import Format, Sndfile

from scipy.signal import firwin, lfilterfrom tempfile import mkstempimport numpy as npimport urllib2import mathimport sysimport reimport os

#For Portuguese Brazilian Speech Recognizer!Lang="pt-BR"

#or for English Speech Recognizer#Lang="en-US"

url = 'https://www.google.com/speech-api/v1/recognize?xjerr=1&client=chromium&lang='+Lang

silence=Trueenv = {}RawRate=8000chunk=1024

#http://en.wikipedia.org/wiki/Vocal_range#Assuming Vocal Range Frequency upper than 75 HzVocalRange = 75.0

#cd, FileNameTmp = mkstemp('TmpSpeechFile.flac')

#Assuming Energy threshold upper than 15 dBThreshold = 15

#10 seconds x 16000 samples/second x ( 16 bits / 8bits/byte ) = 160000 bytes#160000/1024 = +/- 157#157*1024 = 160768TimeoutSignal = 160768

#then 1 second x 16000 = 16000#16000/1024 = 15,625 round to 16#16*1024 = 16384Timeout_NoSpeaking=16384

#normalization for RMS CalcSHORT_NORMALIZE = (1.0/32768.0)

#LastBlock=''

#File Descriptor delivery in AsteriskFD=3

#Open File Descriptorfile=os.fdopen(FD, 'rb')

signal=0

all=[]

while 1:line = sys.stdin.readline().strip()

if line == '':breakkey,data = line.split(':')if key[:4] <> 'agi_':sys.stderr.write("Did not work!\n");sys.stderr.flush()continuekey = key.strip()data = data.strip()if key <> '':env[key] = data

for key in env.keys():sys.stderr.write(" -- %s = %s\n" % (key, env[key]))sys.stderr.flush()

def SendSpeech(File):flac=open(File,"rb").read()os.remove(File)header = {'Content-Type' : 'audio/x-flac; rate=8000'}req = urllib2.Request(url, flac, header)data = urllib2.urlopen(req)find = re.findall('"utterance":(.*),', data.read())#utterancetry:result = find[0].replace('"', '')except:sys.stdout.write("EXEC " + "\"" + "NOOP" + "\" \"" + "speech not recognized ..." + "\"" + "\n")sys.stdout.flush()if result:sys.stdout.write('SET VARIABLE GoogleUtterance "%s"\n'% str(result))sys.stdout.flush()sys.stdout.write("EXEC " + "\"" + "NOOP" + "\" \"" "%s \n"% str(result))sys.stdout.flush()

def Filter(samps):FC = 0.05/(0.5*RawRate)

N = 200a = 1b = firwin(N, cutoff=FC, window='hamming')return lfilter(b, a, samps)

def Pitch(signal):if sys.version_info < (2, 6):crossing =[]for s in signal:crossing.append(s)else:crossing = [math.copysign(1.0, s) for s in signal]#index = find(np.diff(crossing));index = np.nonzero(np.diff(crossing));index=np.array(index)[0].tolist()f0=round(len(index) *RawRate /(2*np.prod(len(signal))))return f0;

def rms(shorts):rms2=0count = len(shorts)/2sum_squares = 0.0for sample in shorts:n = sample * SHORT_NORMALIZEsum_squares += n*nrms2 = math.pow(sum_squares/count,0.5)return rms2 * 1000

def speaking(data):rms_value = rms(data)if rms_value > Threshold:return Trueelse:return False

def VAD(SumFrequency, data2):AVGFrequency = SumFrequency/(Timeout_NoSpeaking+1);if AVGFrequency > VocalRange/2:S=speaking(data2)if S:return True;else:return False;

else:return False;

def RecordSpeech(TimeoutSignal, LastBlock, LastLastBlock):for s in LastLastBlock:

all.append(s)for s in LastBlock:all.append(s)signal=0;while signal <= TimeoutSignal:RawSamps = file.read(Timeout_NoSpeaking)samps = np.fromstring(RawSamps, dtype=np.int16)for s in samps:all.append(s)signal = signal + Timeout_NoSpeaking;#rms_value=rms(samps)Speech=speaking(samps)#sys.stdout.write("EXEC NOOP %s \"\"\"\n"% str(rms_value))#sys.stdout.flush()

#if rms_value > Threshold:if Speech:sys.stdout.write("EXEC " + "\"" + "NOOP" + "\" \"" + "Speech Found ..." + "\" " + "\n")sys.stdout.flush()else:sys.stdout.write("EXEC " + "\"" + "NOOP" + "\" \"" + "End of the Speech..." + "\" " +"\n")sys.stdout.flush()signal=TimeoutSignal+1

def PlayStream (params):sys.stderr.write("STREAM FILE %s \"\"\n" % str(params))sys.stderr.flush()sys.stdout.write("STREAM FILE %s \"\"\n" % str(params))sys.stdout.flush()result = sys.stdin.readline().strip()

sys.stdout.write("EXEC " + "\"" + "NOOP" + "\" \"" + "Hello Waiting For Speech ..." +"\" " + "\n")sys.stdout.flush()

PlayStream("beep");sys.stdout.flush()

while silence:#Input Real-time Data Raw Audio from AsteriskRawSamps = file.read(chunk)samps = np.fromstring(RawSamps, dtype=np.int16)samps2=Filter(samps)Frequency=Pitch(samps2)rms_value=rms(samps)signal = signal + chunk;if (rms_value > Threshold) and (Frequency > VocalRange):silence=False

LastLastBlock=LastBlockLastBlock=sampssys.stdout.write("EXEC " + "\"" + "NOOP" + "\" \"" + "Speech Detected Recording..."+ "\" " + "\n")sys.stdout.flush()if (signal > TimeoutSignal):sys.stdout.write("EXEC " + "\"" + "NOOP" + "\" \"" + "Time Out No Speech Detected..." + "\" " + "\n")sys.stdout.flush()sys.exit()

RecordSpeech(TimeoutSignal, LastBlock, LastLastBlock)

array = np.array(all)

fmt = Format('flac', 'pcm16')nchannels = 1

cd, FileNameTmp = mkstemp('TmpSpeechFile.flac')

# making the file .flacafile = Sndfile(FileNameTmp, 'w', fmt, nchannels, RawRate)

#writing in the fileafile.write_frames(array)

SendSpeech(FileNameTmp)

# FIM ---------- CORTE AQUI -----------------

Creditos: Eng Eder Wander