import scipy
from scipy.io.wavfile import read
import matplotlib.pyplot as plt
import numpy as np
import sys
from utils import calculatefft
fs, data = read(sys.argv[1], mmap=False)
amp, frq = calculatefft(fs, data)
plt.xlabel('frequency')
plt.ylabel('amplitude')
plt.plot(frq, amp)
plt.show()
freqs = np.array(np.exp(np.linspace(np.log(264), np.log(1056), 15)))
round_freqs = np.round(freqs, 0)
amps = amp[round_freqs.astype(np.int64)]
test = np.array([264, 297, 330, 352, 396, 440, 495, 528, 594, 660, 704, 792, 880, 990, 1056])
#T = n/float(fs)
#frq = k/T
T = len(data)/float(fs)
k = test * T
print k
print np.amax(amp)
print np.where(amp == np.amax(amp))
def prepare_harder():
freqs = np.array(np.exp(np.linspace(np.log(264), np.log(2000), 150)))
round_freqs = np.round(freqs, 0)
T = 0.1 #n/fs = 9600/96000
chunk = 9600
round_freqs = round_freqs * T #indeski frekvencija
timedata = [] #sample-ovi, svaki red je 9600 samplova
#prvi ton
fs, data = read('training/trainingG4.wav')
for i in range(0, 3000):
timedata.append(data[i*chunk:i*chunk+chunk])
#drugi ton
fs, data = read('training/trainingF5.wav')
for i in range(0, 3000):
timedata.append(data[i*chunk:i*chunk+chunk])
fs, data = read('training/trainingD6.wav')
for i in range(0, 3000):
timedata.append(data[i*chunk:i*chunk+chunk])
fs, data = read('training/trainingA6.wav')
for i in range(0, 3000):
timedata.append(data[i*chunk:i*chunk+chunk])
timedata = np.array(timedata)
fftdata = []
for row in timedata: #za svaki zvuk se racuna fft
fftdata.append(calculatefft(fs, row)[0])
amps = []
for row in fftdata: #uzimanje odgovarajucih amplituda za frekvencije
#amps.append(row[round_freqs])
amps.append(adaptivelocalmax(row, round_freqs))
y = []
for i in range(0, 3000):
y.append([1, 0, 0, 0])
for i in range(0, 3000):
y.append([0, 1, 0, 0])
for i in range(0, 3000):
y.append([0, 0, 1, 0])
for i in range(0, 3000):
y.append([0, 0, 0, 1])
x = np.array(amps)
y = np.array(y)
return x, y
stream.close() p.terminate() done = True if event.type == pygame.KEYDOWN and event.key == pygame.K_s: pl.orientation = 0 if event.type == pygame.KEYDOWN and event.key == pygame.K_w: pl.orientation = 1 if event.type == pygame.KEYDOWN and event.key == pygame.K_d: pl.orientation = 2 if event.type == pygame.KEYDOWN and event.key == pygame.K_a: pl.orientation = 3 a = record(p, stream) #rms = sqrt(mean(square(a))) a = calculatefft(RATE, a)[0] amptest = [] amptest.append(adaptivelocalmax(a, round_freqs.astype(np.int64))) res = ann.predict(np.array(amptest)) a = np.argmax(res) b = np.amax(res) if b > 0.98: pl.orientation = a pl.move() if(pl.checkApple(ap)): ap.generatePosition()
def prepare_data():
freqs = np.array([264, 297, 330, 352, 396, 440, 495, 528, 594, 660, 704, 792, 880, 990, 1056]) #odredjivanje potrebnih frekvencija
#freqs = np.array(np.exp(np.linspace(np.log(264), np.log(1056), 15)))
chunk = 9600 #96000/10 (sample rate je 96000, uzimam deseti deo sekunde)
T = 0.1 #samples/fs 9600/96000
freqs = freqs * T
round_freqs = np.round(freqs, 0).astype(np.int64)
#data = np.array(data)
#high = np.amax(data)
#data = data/high
#seckanje zvuka od n s, na n zvukova od 1s
print round_freqs
timedata = [] #sample-ovi, svaki red je 9600 samplova
#prvi ton
#fs, data = read(filename, mmap=False)
fs, data = read('training/training440.wav')
for i in range(0, 3000):
timedata.append(data[i*chunk:i*chunk+chunk])
#drugi ton
fs, data = read('training/training660.wav')
for i in range(0, 3000):
timedata.append(data[i*chunk:i*chunk+chunk])
fs, data = read('training/training330.wav')
for i in range(0, 3000):
timedata.append(data[i*chunk:i*chunk+chunk])
fs, data = read('training/training990.wav')
for i in range(0, 3000):
timedata.append(data[i*chunk:i*chunk+chunk])
timedata = np.array(timedata)
fftdata = []
for row in timedata: #za svaki zvuk se racuna fft
fftdata.append(calculatefft(fs, row)[0])
#fftdata = np.array(fftdata)
#fftdata = fftdata/10000000000
amps = []
for row in fftdata: #uzimanje odgovarajucih amplituda za frekvencije
#amps.append(row[round_freqs])
amps.append(localmax(row, round_freqs))
y = []
for i in range(0, 3000):
y.append([1, 0, 0, 0])
for i in range(0, 3000):
y.append([0, 1, 0, 0])
for i in range(0, 3000):
y.append([0, 0, 1, 0])
for i in range(0, 3000):
y.append([0, 0, 0, 1])
x = np.array(amps)
y = np.array(y)
return x, y
test = np.array([264, 297, 330, 352, 396, 440, 495, 528, 594, 660, 704, 792, 880, 990, 1056])
T = 0.1
test = test * T
test = np.round(test, 0)
#freqs = np.array(np.exp(np.linspace(np.log(264), np.log(1056), 15)))
#round_freqs = np.round(freqs, 0)
ann = create_and_train()
amptest = []
fst, datat = read('test/200test.wav', mmap = False)
datat = datat[0:9600]
testfft = calculatefft(fst, datat)[0]
#testfft = testfft/10000000000
#amptest.append(testfft[test.astype(np.int64)])
amptest.append(localmax(testfft, test.astype(np.int64)))
fst, datat = read('test/330test.wav', mmap = False)
datat = datat[0:9600]
testfft = calculatefft(fst, datat)[0]
#testfft = testfft/10000000000
#amptest.append(testfft[test.astype(np.int64)])
amptest.append(localmax(testfft, test.astype(np.int64)))
fst, datat = read('test/400test.wav', mmap = False)
datat = datat[0:9600]
