TensorflowPredictEffnetDiscogs in streaming mode creates empty output on short audio

[dbogdanov]

trafficstars

When used in streaming mode, TensorflowPredictEffnetDiscogs requires 64 second input audio to be able to produce predictions. This is due to a fixed batch size = 64.

• This behavior is not documented.
• Ideally, it should be fixed to allow short audio inputs.