Maximum real-time speed of the model

[eldar81]

I'm creating a real-time pitch detector with the help of CREPE.

If the model is learned on 16 000 Hz data and takes as input 1024 samples, is it true that the maximum real-time updating speed of the model = 1024 / 16 000 = 64ms?

Even if to use step-size smaller than 64ms the will be a delay in this 64ms because model can not process smaller chunk of data.

Am I right? Is there any solution to increase the response speed of the model in real-time prediction?

I am using pyaudio to create audio stream.

[zumpchke]

Did you get anywhere with this?