Update trainSpeakerNet.py

[mst-rajatmishra]

hi,
I Tried making some changes,

1. Introduced logging to replace print statements. This allows for more granular control of the log levels (e.g., INFO, DEBUG, ERROR).
2. Added autocast and GradScaler for mixed precision training in the trainer loop.
3. Used pin_memory=True for DataLoader to speed up data transfer to the GPU.
4. Updated the learning rate decay process to check for test_interval.
5. Improved the comments and clarified some sections of the code.

Thank You,
RAJAT MISHRA