def output(partIdx):
"""Uses the student code to compute the output for test cases."""
outputString = ''
if partIdx == 0: # This is ScaledFFTdB
from assignment1 import scaled_fft_db
r,x = wavfile.read('data/a1_submissionInput.wav')
X = scaled_fft_db(x)
for val in X:
outputString += '%.5f ' % (val)
elif partIdx == 1: # This is PrototypeFilter
from assignment2 import prototype_filter
h = prototype_filter()
# test signal
s = np.loadtxt('data/a2_submissionInput.txt')
r = np.convolve(h, s)[4*512:5*512]/2
for val in r:
outputString += '%.5f ' % val
elif partIdx == 2: # This is SubbandFiltering
from assignment3 import subband_filtering
r,x = wavfile.read('data/a3_submissionInput.wav')
h = np.hanning(512)
X = subband_filtering(x, h)
for val in X:
outputString += '%.5f ' % (val)
elif partIdx == 3: # This is Quantization
from assignment4 import quantization
from parameters import EncoderParameters
params = EncoderParameters(44100, 2, 64)
val_in = np.loadtxt('data/a4_submissionInput.txt')
for r,row in enumerate(val_in):
val = row[0]
scf = row[1]
ba = int(row[2])
QCa = params.table.qca[ba-2]
QCb = params.table.qcb[ba-2]
val = quantization(val, scf, ba, QCa, QCb)
outputString += '%d ' % (val)
return outputString.strip()
def output(partIdx):
"""Uses the student code to compute the output for test cases."""
outputString = ''
if partIdx == 0: # This is ScaledFFTdB
from assignment1 import scaled_fft_db
r, x = wavfile.read('data/a1_submissionInput.wav')
X = scaled_fft_db(x)
for val in X:
outputString += '%.5f ' % (val)
elif partIdx == 1: # This is PrototypeFilter
from assignment2 import prototype_filter
h = prototype_filter()
# test signal
s = np.loadtxt('data/a2_submissionInput.txt')
r = np.convolve(h, s)[4 * 512:5 * 512] / 2
for val in r:
outputString += '%.5f ' % val
elif partIdx == 2: # This is SubbandFiltering
from assignment3 import subband_filtering
r, x = wavfile.read('data/a3_submissionInput.wav')
h = np.hanning(512)
X = subband_filtering(x, h)
for val in X:
outputString += '%.5f ' % (val)
elif partIdx == 3: # This is Quantization
from assignment4 import quantization
from parameters import EncoderParameters
params = EncoderParameters(44100, 2, 64)
val_in = np.loadtxt('data/a4_submissionInput.txt')
for r, row in enumerate(val_in):
val = row[0]
scf = row[1]
ba = int(row[2])
QCa = params.table.qca[ba - 2]
QCb = params.table.qcb[ba - 2]
val = quantization(val, scf, ba, QCa, QCb)
outputString += '%d ' % (val)
return outputString.strip()
def output(partIdx):
outputString = ""
if partIdx == "1": # This is ScaledFFTdB
from assignment1 import scaled_fft_db
r, x = wavfile.read("data/a1_submissionInput.wav")
X = scaled_fft_db(x)
for val in X:
outputString += "%.5f " % (val)
elif partIdx == "2": # This is PrototypeFilter
from assignment2 import prototype_filter
h = prototype_filter()
# test signal
s = np.loadtxt("data/a2_submissionInput.txt")
r = np.convolve(h, s)[4 * 512:5 * 512] / 2
for val in r:
outputString += "%.5f " % val
elif partIdx == "3": # This is SubbandFiltering
from assignment3 import subband_filtering
r, x = wavfile.read("data/a3_submissionInput.wav")
h = np.hanning(512)
X = subband_filtering(x, h)
for val in X:
outputString += "%.5f " % (val)
elif partIdx == "4": # This is Quantization
from assignment4 import quantization
from parameters import EncoderParameters
params = EncoderParameters(44100, 2, 64)
val_in = np.loadtxt("data/a4_submissionInput.txt")
for r, row in enumerate(val_in):
val = row[0]
scf = row[1]
ba = int(row[2])
QCa = params.table.qca[ba - 2]
QCb = params.table.qcb[ba - 2]
val = quantization(val, scf, ba, QCa, QCb)
outputString += "%d " % (val)
else:
print("Unknown assigment part number")
if len(outputString) > 0:
fileName = "res%s.txt" % partIdx
with open(fileName, "w") as f:
f.write(outputString.strip())
print("You can now submit the file " + fileName)
else:
print(
"there was an error with the computation. Please check your code")
def output(partIdx):
outputString = ''
if partIdx == '1': # This is ScaledFFTdB
from assignment1 import scaled_fft_db
r,x = wavfile.read('data/a1_submissionInput.wav')
X = scaled_fft_db(x)
for val in X:
outputString += '%.5f ' % (val)
elif partIdx == '2': # This is PrototypeFilter
from assignment2 import prototype_filter
h = prototype_filter()
# test signal
s = np.loadtxt('data/a2_submissionInput.txt')
r = np.convolve(h, s)[4*512:5*512]/2
for val in r:
outputString += '%.5f ' % val
elif partIdx == '3': # This is SubbandFiltering
from assignment3 import subband_filtering
r,x = wavfile.read('data/a3_submissionInput.wav')
h = np.hanning(512)
X = subband_filtering(x, h)
for val in X:
outputString += '%.5f ' % (val)
elif partIdx == '4': # This is Quantization
from assignment4 import quantization
from parameters import EncoderParameters
params = EncoderParameters(44100, 2, 64)
val_in = np.loadtxt('data/a4_submissionInput.txt')
for r,row in enumerate(val_in):
val = row[0]
scf = row[1]
ba = int(row[2])
QCa = params.table.qca[ba-2]
QCb = params.table.qcb[ba-2]
val = quantization(val, scf, ba, QCa, QCb)
outputString += '%d ' % (val)
else:
print "Unknown assigment part number"
if len(outputString) > 0:
fileName = "res%s.txt" % partIdx;
with open(fileName, "w") as f:
f.write(outputString.strip())
print "You can now submit the file " + fileName
else:
print "there was an error with the computation. Please check your code"
def model1(samples, params, sfindices):
"""Psychoacoustic model as described in ISO/IEC 11172-3, Annex D.1."""
table = params.table
X = assignment1.scaled_fft_db(samples)
scf = table.scalefactor[sfindices]
subband_spl = np.zeros(N_SUBBANDS)
for sb in range(N_SUBBANDS):
subband_spl[sb] = np.max(X[1 + sb * SUB_SIZE:1 + sb * SUB_SIZE +
SUB_SIZE])
subband_spl[sb] = np.maximum(subband_spl[sb],
20 * np.log10(scf[0, sb] * 32768) - 10)
peaks = []
for i in range(3, FFT_SIZE / 2 - 6):
if X[i] >= X[i + 1] and X[i] > X[i - 1]:
peaks.append(i)
# determining tonal and non-tonal components
tonal = TonalComponents(X)
tonal.flag[0:3] = IGNORE
for k in peaks:
is_tonal = True
if k > 2 and k < 63:
testj = [-2, 2]
elif k >= 63 and k < 127:
testj = [-3, -2, 2, 3]
else:
testj = [-6, -5, -4, -3, -2, 2, 3, 4, 5, 6]
for j in testj:
if tonal.spl[k] - tonal.spl[k + j] < 7:
is_tonal = False
break
if is_tonal:
tonal.spl[k] = add_db(tonal.spl[k - 1:k + 2])
tonal.flag[k + np.arange(testj[0], testj[-1] + 1)] = IGNORE
tonal.flag[k] = TONE
tonal.tonecomps.append(k)
# non-tonal components for each critical band
for i in range(table.cbnum - 1):
weight = 0.0
msum = DBMIN
for j in range(table.cbound[i], table.cbound[i + 1]):
if tonal.flag[i] == UNSET:
msum = add_db((tonal.spl[j], msum))
weight += np.power(
10, tonal.spl[j] / 10) * (table.bark[table.map[j]] - i)
if msum > DBMIN:
index = weight / np.power(10, msum / 10.0)
center = table.cbound[i] + np.int(
index * (table.cbound[i + 1] - table.cbound[i]))
if tonal.flag[center] == TONE:
center += 1
tonal.flag[center] = NOISE
tonal.spl[center] = msum
tonal.noisecomps.append(center)
# decimation of tonal and non-tonal components
# under the threshold in quiet
for i in range(len(tonal.tonecomps)):
if i >= len(tonal.tonecomps):
break
k = tonal.tonecomps[i]
if tonal.spl[k] < table.hear[table.map[k]]:
tonal.tonecomps.pop(i)
tonal.flag[k] = IGNORE
i -= 1
for i in range(len(tonal.noisecomps)):
if i >= len(tonal.noisecomps):
break
k = tonal.noisecomps[i]
if tonal.spl[k] < table.hear[table.map[k]]:
tonal.noisecomps.pop(i)
tonal.flag[k] = IGNORE
i -= 1
# decimation of tonal components closer than 0.5 Bark
for i in range(len(tonal.tonecomps) - 1):
if i >= len(tonal.tonecomps) - 1:
break
this = tonal.tonecomps[i]
next = tonal.tonecomps[i + 1]
if table.bark[table.map[this]] - table.bark[table.map[next]] < 0.5:
if tonal.spl[this] > tonal.spl[next]:
tonal.flag[next] = IGNORE
tonal.tonecomps.remove(next)
else:
tonal.flag[this] = IGNORE
tonal.tonecomps.remove(this)
# individual masking thresholds
masking_tonal = []
masking_noise = []
for i in range(table.subsize):
masking_tonal.append(())
zi = table.bark[i]
for j in tonal.tonecomps:
zj = table.bark[table.map[j]]
dz = zi - zj
if dz >= -3 and dz <= 8:
avtm = -1.525 - 0.275 * zj - 4.5
if dz >= -3 and dz < -1:
vf = 17 * (dz + 1) - (0.4 * X[j] + 6)
elif dz >= -1 and dz < 0:
vf = dz * (0.4 * X[j] + 6)
elif dz >= 0 and dz < 1:
vf = -17 * dz
else:
vf = -(dz - 1) * (17 - 0.15 * X[j]) - 17
masking_tonal[i] += (X[j] + vf + avtm, )
for i in range(table.subsize):
masking_noise.append(())
zi = table.bark[i]
for j in tonal.noisecomps:
zj = table.bark[table.map[j]]
dz = zi - zj
if dz >= -3 and dz <= 8:
avnm = -1.525 - 0.175 * zj - 0.5
if dz >= -3 and dz < -1:
vf = 17 * (dz + 1) - (0.4 * X[j] + 6)
elif dz >= -1 and dz < 0:
vf = dz * (0.4 * X[j] + 6)
elif dz >= 0 and dz < 1:
vf = -17 * dz
else:
vf = -(dz - 1) * (17 - 0.15 * X[j]) - 17
masking_noise[i] += (X[j] + vf + avnm, )
# global masking thresholds
masking_global = []
for i in range(table.subsize):
maskers = (table.hear[i], ) + masking_tonal[i] + masking_noise[i]
masking_global.append(add_db(maskers))
# minimum masking thresholds
mask = np.zeros(N_SUBBANDS)
for sb in range(N_SUBBANDS):
first = table.map[sb * SUB_SIZE]
after_last = table.map[(sb + 1) * SUB_SIZE - 1] + 1
mask[sb] = np.min(masking_global[first:after_last])
# signal-to-mask ratio for each subband
smr = subband_spl - mask
subband_bit_allocation = smr_bit_allocation(params, smr)
return subband_bit_allocation
def model1(samples, params, sfindices):
"""Psychoacoustic model as described in ISO/IEC 11172-3, Annex D.1."""
table = params.table
X = assignment1.scaled_fft_db(samples)
scf = table.scalefactor[sfindices]
subband_spl = np.zeros(N_SUBBANDS)
for sb in range(N_SUBBANDS):
subband_spl[sb] = np.max(X[1 + sb * SUB_SIZE : 1 + sb * SUB_SIZE + SUB_SIZE])
subband_spl[sb] = np.maximum(subband_spl[sb], 20 * np.log10(scf[0,sb] * 32768) - 10)
peaks = []
for i in range(3, FFT_SIZE / 2 - 6):
if X[i]>=X[i+1] and X[i]>X[i-1]:
peaks.append(i)
#determining tonal and non-tonal components
tonal = TonalComponents(X)
tonal.flag[0:3] = IGNORE
for k in peaks:
is_tonal = True
if k > 2 and k < 63:
testj = [-2,2]
elif k >= 63 and k < 127:
testj = [-3,-2,2,3]
else:
testj = [-6,-5,-4,-3,-2,2,3,4,5,6]
for j in testj:
if tonal.spl[k] - tonal.spl[k+j] < 7:
is_tonal = False
break
if is_tonal:
tonal.spl[k] = add_db(tonal.spl[k-1:k+2])
tonal.flag[k+np.arange(testj[0], testj[-1] + 1)] = IGNORE
tonal.flag[k] = TONE
tonal.tonecomps.append(k)
#non-tonal components for each critical band
for i in range(table.cbnum - 1):
weight = 0.0
msum = DBMIN
for j in range(table.cbound[i], table.cbound[i+1]):
if tonal.flag[i] == UNSET:
msum = add_db((tonal.spl[j], msum))
weight += np.power(10, tonal.spl[j] / 10) * (table.bark[table.map[j]] - i)
if msum > DBMIN:
index = weight/np.power(10, msum / 10.0)
center = table.cbound[i] + np.int(index * (table.cbound[i+1] - table.cbound[i]))
if tonal.flag[center] == TONE:
center += 1
tonal.flag[center] = NOISE
tonal.spl[center] = msum
tonal.noisecomps.append(center)
#decimation of tonal and non-tonal components
#under the threshold in quiet
for i in range(len(tonal.tonecomps)):
if i >= len(tonal.tonecomps):
break
k = tonal.tonecomps[i]
if tonal.spl[k] < table.hear[table.map[k]]:
tonal.tonecomps.pop(i)
tonal.flag[k] = IGNORE
i -= 1
for i in range(len(tonal.noisecomps)):
if i >= len(tonal.noisecomps):
break
k = tonal.noisecomps[i]
if tonal.spl[k] < table.hear[table.map[k]]:
tonal.noisecomps.pop(i)
tonal.flag[k] = IGNORE
i -= 1
#decimation of tonal components closer than 0.5 Bark
for i in range(len(tonal.tonecomps) -1 ):
if i >= len(tonal.tonecomps) -1:
break
this = tonal.tonecomps[i]
next = tonal.tonecomps[i+1]
if table.bark[table.map[this]] - table.bark[table.map[next]] < 0.5:
if tonal.spl[this]>tonal.spl[next]:
tonal.flag[next] = IGNORE
tonal.tonecomps.remove(next)
else:
tonal.flag[this] = IGNORE
tonal.tonecomps.remove(this)
#individual masking thresholds
masking_tonal = []
masking_noise = []
for i in range(table.subsize):
masking_tonal.append(())
zi = table.bark[i]
for j in tonal.tonecomps:
zj = table.bark[table.map[j]]
dz = zi - zj
if dz >= -3 and dz <= 8:
avtm = -1.525 - 0.275 * zj - 4.5
if dz >= -3 and dz < -1:
vf = 17 * (dz + 1) - (0.4 * X[j] + 6)
elif dz >= -1 and dz < 0:
vf = dz * (0.4 * X[j] + 6)
elif dz >= 0 and dz < 1:
vf = -17 * dz
else:
vf = -(dz - 1) * (17 - 0.15 * X[j]) - 17
masking_tonal[i] += (X[j] + vf + avtm,)
for i in range(table.subsize):
masking_noise.append(())
zi = table.bark[i]
for j in tonal.noisecomps:
zj = table.bark[table.map[j]]
dz = zi - zj
if dz >= -3 and dz <= 8:
avnm = -1.525 - 0.175 * zj - 0.5
if dz >= -3 and dz < -1:
vf = 17 * (dz + 1) - (0.4 * X[j] + 6)
elif dz >= -1 and dz < 0:
vf = dz * (0.4 * X[j] + 6)
elif dz >= 0 and dz < 1:
vf = -17 * dz
else:
vf = -(dz - 1) * (17 - 0.15 * X[j]) - 17
masking_noise[i] += (X[j] + vf + avnm,)
#global masking thresholds
masking_global = []
for i in range(table.subsize):
maskers = (table.hear[i],) + masking_tonal[i] + masking_noise[i]
masking_global.append(add_db(maskers))
#minimum masking thresholds
mask = np.zeros(N_SUBBANDS)
for sb in range(N_SUBBANDS):
first = table.map[sb * SUB_SIZE]
after_last = table.map[(sb + 1) * SUB_SIZE - 1] + 1
mask[sb] = np.min(masking_global[first:after_last])
#signal-to-mask ratio for each subband
smr = subband_spl - mask
subband_bit_allocation = smr_bit_allocation(params, smr)
return subband_bit_allocation