pitch

Pitch/PitchTest.py

@@ -0,0 +1,46 @@
+import wave
+import numpy as np
+import pylab as pl
+
+# read wave file and get parameters.
+fw = wave.open('../sounds/a.wav','rb')
+params = fw.getparams()
+print(params)
+nchannels, sampwidth, framerate, nframes = params[:4]
+strData = fw.readframes(nframes)
+waveData = np.fromstring(strData, dtype=np.int16)
+waveData = waveData*1.0/max(abs(waveData))  # normalization
+fw.close()
+
+# plot the wave
+time = np.arange(0, len(waveData)) * (1.0 / framerate)
+
+index1 = 10000.0 / framerate
+index2 = 10512.0 / framerate
+index3 = 15000.0 / framerate
+index4 = 15512.0 / framerate
+
+pl.subplot(311)
+pl.title("pitch")
+pl.plot(time, waveData)
+pl.plot([index1,index1],[-1,1],'r')
+pl.plot([index2,index2],[-1,1],'r')
+pl.plot([index3,index3],[-1,1],'g')
+pl.plot([index4,index4],[-1,1],'g')
+pl.xlabel("time (seconds)")
+pl.ylabel("Amplitude")
+
+pl.subplot(312)
+pl.plot(np.arange(512),waveData[10000:10512],'r')
+pl.plot([59,59],[-1,1],'b')
+pl.plot([169,169],[-1,1],'b')
+print(1/( (169-59)*1.0/framerate ))
+pl.xlabel("index in 1 frame")
+pl.ylabel("Amplitude")
+
+pl.subplot(313)
+pl.plot(np.arange(512),waveData[15000:15512],'g')
+pl.xlabel("index in 1 frame")
+pl.ylabel("Amplitude")
+pl.savefig("pitch.png")
+pl.show()

Pitch/PitchTracking.py

@@ -0,0 +1,75 @@
+import math
+import numpy as np
+
+# Orignal auto-correlation function(ACF)
+def ACF(frame):
+    flen = len(frame)
+    acf = np.zeros(flen)
+    for i in range(flen):
+        acf[i] = np.sum(frame[i:flen]*frame[0:flen-i])
+    return acf
+
+# ACF with weight
+def ACF2(frame):
+    flen = len(frame)
+    acf = np.zeros(flen)
+    for i in range(flen):
+        acf[i] = np.sum(frame[i:flen]*frame[0:flen-i])/(flen-i)
+    return acf
+
+# ACF to half frame length
+def ACF3(frame):
+    flen = len(frame)
+    acf = np.zeros(flen/2)
+    for i in range(flen/2):
+        acf[i] = np.sum(frame[i:flen]*frame[0:flen-i])
+    return acf
+
+# normalized squared difference function(NSDF)
+def NSDF(frame):
+    flen = len(frame)
+    nsdf = np.zeros(flen)
+    for i in range(flen):
+        s1 = np.sum(frame[i:flen]*frame[0:flen-i])
+        s2 = np.sum(frame[i:flen]*frame[i:flen])
+        s3 = np.sum(frame[0:flen-i]*frame[0:flen-i])
+        nsdf[i] = 2.0*s1/(s2+s3)
+    return nsdf
+
+# AMDF (average magnitude difference function) 
+def AMDF(frame):
+    flen = len(frame)
+    amdf = np.zeros(flen)
+    for i in range(flen):
+        amdf[i] = -np.sum(np.abs(frame[i:flen]-frame[0:flen-i]))  # to adjust to ACF, I use the -AMDF
+    return amdf
+
+# AMDF with weight
+def AMDF2(frame):
+    flen = len(frame)
+    amdf = np.zeros(flen)
+    for i in range(flen):
+        amdf[i] = -np.sum(np.abs(frame[i:flen]-frame[0:flen-i]))/(flen-i)  # to adjust to ACF, I use the -AMDF
+    return amdf
+
+# AMDF to half frame length
+def AMDF3(frame):
+    flen = len(frame)
+    amdf = np.zeros(flen/2)
+    for i in range(flen/2):
+        amdf[i] = -np.sum(np.abs(frame[i:flen]-frame[0:flen-i]))  # to adjust to ACF, I use the -AMDF
+    return amdf
+
+# Pitch Tracking
+def PitchTrack(waveData,frameRate,frameSize,overLap,acfmethod):
+    wlen = len(waveData)
+    step = frameSize - overLap
+    frameNum = int(math.ceil(wlen*1.0/step))
+    pitchtrack = np.zeros(frameNum)
+    for i in range(frameNum):
+        #acf = acfmethod(waveData[i*step : i*step+frameSize])  # Bug
+        acf = acfmethod(waveData[int(i*step):int(i*step+frameSize)])
+        acf[0:30] = np.min(acf)
+        acfmax = np.argmax(acf)
+        pitchtrack[i] = frameRate*1.0/acfmax
+    return pitchtrack

Pitch/PitchTrackingTest.py

@@ -0,0 +1,59 @@
+import wave
+import numpy as np
+import pylab as pl
+import PitchTracking as pt
+
+# read wave file and get parameters.
+fw = wave.open('../sounds/aeiou.wav','rb')
+params = fw.getparams()
+print(params)
+nchannels, sampwidth, framerate, nframes = params[:4]
+strData = fw.readframes(nframes)
+waveData = np.fromstring(strData, dtype=np.int16)
+waveData = waveData*1.0/max(abs(waveData))  # normalization
+fw.close()
+
+# plot the wave
+time = np.arange(0, len(waveData)) * (1.0 / framerate)
+
+frameSize = 512
+overLap = frameSize/2
+idx1 = 10000
+idx2 = idx1+frameSize
+index1 = idx1*1.0 / framerate
+index2 = idx2*1.0 / framerate
+acf = pt.ACF(waveData[idx1:idx2])
+acf[0:10] = -acf[0]
+acfmax = np.argmax(acf)
+print(acfmax)
+print(framerate*1.0/acfmax)
+
+pl.subplot(411)
+pl.title("pitchTrack")
+pl.plot(time, waveData)
+pl.plot([index1,index1],[-1,1],'r')
+pl.plot([index2,index2],[-1,1],'r')
+pl.xlabel("time (seconds)")
+pl.ylabel("Amplitude")
+
+pl.subplot(412)
+pl.plot(np.arange(frameSize),waveData[idx1:idx2],'r')
+pl.xlabel("index in 1 frame")
+pl.ylabel("Amplitude")
+
+pl.subplot(413)
+pl.plot(np.arange(frameSize),acf,'g')
+pl.xlabel("index in 1 frame")
+pl.ylabel("ACF")
+
+# pitch tracking
+acfmethod = pt.ACF
+pitchtrack = pt.PitchTrack(waveData, framerate, frameSize, overLap, acfmethod)
+xpt = np.arange(0, len(pitchtrack)) *( len(waveData) *1.0/ len(pitchtrack) / framerate )
+pl.subplot(414)
+pl.plot(xpt,pitchtrack,'-*')
+pl.xlabel('time (seconds)')
+pl.ylabel('Frequency (Hz)')
+
+#pl.savefig("pitchTrack.png")
+pl.show()