Recognize Pitch

The pitch of a signal is an extremely important descriptor for signals like speech or musical instruments. Modern machine learning techniques have consistently improved the reliability of this operation.

Recognize the pitch of a monophonic signal using PitchRecognize.

In[1]:=1

✖

Out[1]=1

In[2]:=2

✖

Out[2]=2

In[3]:=3

✖

Out[3]=3

Use a neural network–based method.

In[4]:=4

✖

Out[4]=4

You can also import and use the CREPE pitch recognition neural net from the Wolfram Neural Net Repository. A HiddenMarkovProcess can be used to interpret the output of the network into a sequence of frequency estimates and their confidence.

In[5]:=5

✖

Out[5]=5

The network was trained to predict an estimate of the pitch as a probability distribution on a set of 360 logarithmic pitch classes. You can define a utility function to interpolate between the class predictions provided by the network as well as a function that outputs the recognized frequency and its confidence.

In[6]:=6

✖

In[7]:=7

✖

In[8]:=8

✖

Recognize the pitch and compute the corresponding confidence.

In[9]:=9

✖

Out[9]=9

Plot the recognized frequency with the confidence mapped to the color.

In[10]:=10

✖

Out[10]=10