def start():
speech = Pygsr()
# duration in seconds
speech.record(4)
output = "log.csv"
input = "audio.flac"
if input[-5:] != '.flac':
input += '*.flac'
files = glob.glob(input)
for flac in files:
print "opening %s:" % flac
valid_result = False
tries = 0
with open(flac, 'rb') as f:
result = GoogleSpeechAPI(f)
print "Audio is %.03f seconds long" % result.length
f.seek(0)
with Timer() as t:
result.start()
print "Result took %.03f sec" % t.interval
print result.result
out = ''.join(result.result)
split3 = ""
try:
split1 = [x.strip() for x in out.split(':')]
split2 = split1[3]
split3 = [x.strip() for x in split2.split(',')]
#[1:-1] is to remove double quotes from string
print split3[0][1:-1]
return split3[0][1:-1]
except Exception, e:
print "Exception"
def witAI():
#loading pygsr Python google speach recognition
speech = Pygsr()
raw_input("Ready ?")
speech.record(5) # duration in seconds (3)
try:
phrase, complete_response = speech.speech_to_text(
'en_US') # select the language
except:
phrase = "Tell me a joke !"
print phrase
phrase = phrase.strip().replace(" ", "%20").encode(
'ascii') #if not ascii, curl crashes
#Wit.ai curl URL
curl_url = "https://api.wit.ai/message?q=%s" % phrase
#Auth headers
curl_header = ["Authorization: Bearer YW3P2YITCYYXGHVLMIE7R7G7BBJODBG4"]
#debug
print curl_url
answer = get_curl(curl_url, curl_header)
print answer
result = json.loads(answer) #parse answer
return result
def start():
speech = Pygsr()
# duration in seconds
speech.record(4)
output = "log.csv"
input = "audio.flac"
if input[-5:] != '.flac':
input += '*.flac'
files = glob.glob(input)
for flac in files:
print "opening %s:" % flac
valid_result = False
tries = 0
with open(flac, 'rb') as f:
result = GoogleSpeechAPI(f)
print "Audio is %.03f seconds long" % result.length
f.seek(0)
with Timer() as t:
result.start()
print "Result took %.03f sec" % t.interval
print result.result
out = ''.join(result.result)
split3 = ""
try:
split1 = [x.strip() for x in out.split(':')]
split2 = split1[3]
split3 = [x.strip() for x in split2.split(',')]
#[1:-1] is to remove double quotes from string
print split3[0][1:-1]
return split3[0][1:-1]
except Exception, e:
print "Exception"
def srprocess(self,threadname):
speech = Pygsr()
speech.record(3)
try:
phrase, complete_response = speech.speech_to_text('ar_AE')
except:
phrase = ''
global do_flage
do_flage = False
self.save_text(phrase+' ')
class HearingModule:
def __init__(self):
self.speech = Pygsr()
def listen(self, t):
self.speech.record(t)
phrase, complete_response = self.speech.sepeech_to_text('en_EN')
return phrase
def connected(self):
if self.speech:
return True
return False
class Recorder:
def __init__(self):
global logger
logger = logging.getLogger(__name__)
self.speech = Pygsr()
def record_command(self):
''' Records audio and sends it to google to translate to text.
'''
self.speech.record(settings.RECORD_LENGTH)
result = self.speech.speech_to_text()
line = None
if result:
line = result[0].lower()
if not line:
logger.warn('No command recorded.')
return None
logger.info(line)
command = Command(line)
logger.info(command)
return command
def witAI():
#loading pygsr Python google speach recognition
speech = Pygsr()
raw_input("Ready ?")
speech.record(5) # duration in seconds (3)
try:
phrase, complete_response = speech.speech_to_text('en_US') # select the language
except:
phrase = "Tell me a joke !"
print phrase
phrase = phrase.strip().replace(" ","%20").encode('ascii') #if not ascii, curl crashes
#Wit.ai curl URL
curl_url = "https://api.wit.ai/message?q=%s"%phrase
#Auth headers
curl_header = ["Authorization: Bearer YW3P2YITCYYXGHVLMIE7R7G7BBJODBG4"]
#debug
print curl_url
answer = get_curl(curl_url,curl_header)
print answer
result = json.loads(answer) #parse answer
return result
def listen(self):
speech = Pygsr()
speech.record(2)
phrase, complete_response = speech.speech_to_text('en_IN')
print phrase
def _voice_input(self, duration):
speech = Pygsr()
speech.record(duration)
phrase, complete_response = speech.speech_to_text(self.lang)
return phrase
from pygsr import Pygsr
import sys
import os
def set_proc_name(newname):
from ctypes import cdll, byref, create_string_buffer
libc = cdll.LoadLibrary('libc.so.6')
buff = create_string_buffer(len(newname)+1)
buff.value = newname
libc.prctl(15, byref(buff), 0, 0, 0)
def get_proc_name():
from ctypes import cdll, byref, create_string_buffer
libc = cdll.LoadLibrary('libc.so.6')
buff = create_string_buffer(128)
# 16 == PR_GET_NAME from <linux/prctl.h>
libc.prctl(16, byref(buff), 0, 0, 0)
return buff.value
set_proc_name('Mwave_gspeech')
speech = Pygsr()
speech.record(2) # duration in seconds (3)
phrase, complete_response = speech.speech_to_text('en-US') # select the language
os.remove('audio')
os.remove('audio.flac')
print phrase
from pygsr import Pygsr
speech = Pygsr()
# duration in seconds (3)
speech.record(3)
# select the language and obtain the result
phrase, complete_response = speech.speech_to_text('es_ES')
print phrase
f.write(opener.open(request).read())
f.close()
Popen(['mplayer', 'data.mp3', '-really-quiet']).wait()
#os.system('mplayer -ao alsa -noconsolecontrols data.mp3')
if __name__ == '__main__':
x = Record()
x.setup()
recorded = False
response = None
while response is not 'exit':
x.read()
rms = audioop.rms(x.data, 2)
print rms
if rms > x.threshold:
speech = Pygsr()
speech.record(5)
phrase, complete_response = speech.speech_to_text('en_EN')
response = x.custom(phrase)
if response == False:
response = x.Wolfram(phrase)
if response == False:
response = x.cleverbot(phrase)
print('PHRASEPHRASEPHRASE')
print(phrase)
print(response)
x.speak(response)
recorded = True
rms = 0
x.setup()
#!/usr/bin/python
from pygsr import Pygsr
speech = Pygsr()
speech.record(3)
phrase, complete_response = speech.speech_to_text('de_DE')
print phrase
from pygsr import Pygsr
speech = Pygsr()
speech.record(3) # duration in seconds (3)
phrase, complete_response = speech.speech_to_text(
'es_ES') # select the language
print phrase
from pygsr import Pygsr
speech = Pygsr()
# duration in seconds (3)
speech.record(3)
# select the language and obtain the result
phrase, complete_response = speech.speech_to_text("es_ES")
print phrase
def __init__(self): Pygsr.__init__(self) self.active = False self.count_silence = 0 return
from pygsr import Pygsr
speech = Pygsr()
# duration in seconds
speech.record(3)
# select the language
(phrase, complete_response) = speech.speech_to_text('en_US')
print(phrase)
#!/usr/bin python
#coding: utf-8
from pygsr import Pygsr
speech = Pygsr()
speech.record(3) # duration in seconds (3)
response = speech.speech_to_text('es_ES') # select the language
print response
def __init__(self):
global logger
logger = logging.getLogger(__name__)
self.speech = Pygsr()
def listen(self):
speech = Pygsr()
speech.record(2)
phrase, complete_response = speech.speech_to_text('en_IN')
print phrase
def __init__(self):
self.speech = Pygsr()
def main(wavFileName):
########################################################################################################################
#wavFileName = "/Users/toine/Documents/speech_recognition/sound/sample/test.wav"
wavFile = wave.open(wavFileName)
(nchannels, sampwidth, framerate, nframes, comptype, compname) = wavFile.getparams()
frames = wavFile.readframes(-1)
npFrames = np.fromstring(frames, "Int16")
########################################################################################################################
## compute the spectrogram
## make sure FFT size is not too big for good accuracy
nFft = 64
nOverlap = 32
fftWindow = nFft - nOverlap
specgramFramerate = framerate / (fftWindow)
##TODO: check if this is needed
## pad the input for perfect FFT match
## npFrames = np.r_[npFrames, np.zeros(nFft - nframes % nFft)]
## spectrogram, return (Pxx, freqs, bins, im)
# bins are the time points the spectrogram is calculated over
# freqs is an array of frequencies
# Pxx is an array of shape (len(times), len(freqs)) of power
# im is a AxesImage instance
(Pxx, freqs, bins, im) = plt.specgram(npFrames, Fs=framerate, NFFT=nFft, noverlap=nOverlap)
#plt.show()
plt.clf()
########################################################################################################################
## extract the voice frequencies
## voice frequency range, from 300Hz to 3500Hz
# create a mask vector with these frequency taken from B
# sum over the voice frequency range, voiceArray is 0's, but 1 when in voice frequency range
f300Ind = lib.overflow(freqs, 300)
f3500Ind = lib.overflow(freqs, 3500)
voiceArray = np.zeros(len(freqs))
voiceArray[f300Ind:f3500Ind] = 1
## dot product of the specgram
voiceFreq = np.transpose(np.dot(np.transpose(Pxx), voiceArray))
########################################################################################################################
## compute the interesting minimums based on minimums and threshold
#TODO: consider using the mlab/numpy function
histData = plt.hist(voiceFreq, bins=100, range=(min(voiceFreq), np.mean(voiceFreq)))
#plt.show()
plt.clf()
overflowPercent = 0.7
overflowIndex = lib.overflow_hist(histData[0], overflowPercent)
overflowValue = histData[1][overflowIndex]
## smooth the curve to find the minimums
voiceFreqSmooth = lib.smooth(voiceFreq, 128)
minimums = np.r_[True, voiceFreqSmooth[1:] < voiceFreqSmooth[:-1]] & \
np.r_[voiceFreqSmooth[:-1] < voiceFreqSmooth[1:], True]
##TODO: change name
## create the array of cutting points, points are local minimums under the histogram threshold
cutPoints = np.where(minimums & (voiceFreqSmooth < overflowValue))[0]
########################################################################################################################
## filter the minimums by roughly selecting one every 5 seconds
# on npFrames, 5 sec = framerate * 5
# on voiceFreq, framerate -> framerate/32
avgSec = 3
cutPointsNSec = [0]
for pt in cutPoints:
pt *= fftWindow # convert cutPointsThres to npFrames framerate by multiplying with fftWindow
if (pt - cutPointsNSec[-1]) > (framerate * avgSec): # subtract the last value
cutPointsNSec.append(pt)
########################################################################################################################
## create the cuts as additional files
cutPointsNSecInSec = [(x / framerate) for x in cutPointsNSec]
timestamp = []
timestampNFrames = []
for item1, item2 in lib.pairwise(cutPointsNSec, fillvalue=0):
timestamp.append((item1, item2))
timestampNFrames.append(item2 - item1)
# geenrate the extension to the filename, e.g. filename.X_Y.wav for a cut from seconds X to Y
addExtension = []
timestampInSec = []
for item1, item2 in lib.pairwise(cutPointsNSecInSec, fillvalue="end"):
tmp = str(item1) + "_" + str(item2)
timestampInSec.append((item1, item2))
addExtension.append(tmp)
logger = logging.getLogger(__name__)
logger.debug("%s %s %s", timestamp, timestampNFrames, addExtension)
logger.debug("%s %s %s", len(timestamp), len(timestampNFrames), len(addExtension))
## test on 1 file first
#for (cutExt, cutTime, cutFrame) in zip(timestamp, timestampNFrames, addExtension):
totalRes = []
TESTINDEX = 6
#TODO: take care of the last index, when cutPointNSecInSec is "end"
for TESTINDEX in range(len(timestamp)-1):
#TODO: make a lib function out of that
splitName = path.basename(wavFileName).split(".")
filename = path.dirname(wavFileName) + "/" + splitName[0] + "." + addExtension[TESTINDEX] + "." + splitName[1]
wavChunk = wave.open(filename, "w")
wavChunk.setparams((nchannels, sampwidth, framerate, timestampNFrames[TESTINDEX], comptype, compname))
wavChunk.writeframes(npFrames[timestamp[TESTINDEX][0]:timestamp[TESTINDEX][1]].tostring())
wavChunk.close()
pygsr = Pygsr(filename)
pygsr.convert()
res = pygsr.speech_to_text("en", indx=TESTINDEX)
totalRes.append(res)
logger.debug("%s %s %s", TESTINDEX, addExtension[TESTINDEX], timestamp[TESTINDEX])
h1 = str(datetime.timedelta(seconds=timestampInSec[TESTINDEX][0]))+",200"
h2 = str(datetime.timedelta(seconds=timestampInSec[TESTINDEX][1]-1))+",800"
logger.info("%s", TESTINDEX)
logger.info("%s --> %s", h1, h2)
logger.info("%s", res)
logger.info("")
#logger.debug("this should not appear in the srt file")
logger.debug("%s", totalRes)
return 1
