Do i need a filter for the Input for LSTMs?

Hi everyone,
im working on an LSTM which gets vibration signal of a mechanical part to predict the remaining useful lifetime. Since the signal is very noisy, i’m not sure if the LSTM is able to learn the data properly. Is it necessary to filter noisy inputs for LSTMs? How is the approach in other fields, where the signal is noisy as well, like for voice signal processing?
I wanted to use the scipy.savgol_filter for the problem.!
Thanks in advance.

blue is the original signal, orange is the filtered signal
The signal is normalized for the training.