Hello,
I’m using TensorFlow for speech/voice generation with 5 neural nets. While capturing wave in and playing the buffers for wave out the speech is processed by the 5 neural nets. This I do with model.predict. I have problems with running the neural nets real-time. The first one or two minutes or so everything works correctly. But then there’s a halt/interruption for 20 seconds and after that the processing goes on. So this causes dropped audio frames. I’m using Keras/TensorFlow 2.10.0 for Delphi (Keras4Delphi). And I’m running the neural nets on the GPU. Please is there someone who knows how I can solve this issue?
Thanks,
Roelof