Is a GPU required to run wav2letter?

[nchp89]

Question

My lab is having a VM set up on a remote machine to decode audio files. From reading the documentation, it seems like a GPU is only required if using the CUDA backend. However, one of the IT guys setting up wav2letter expressed concern because CUDA libraries are required for installation. Is a GPU required to run wav2letter even if one chooses not to use the CUDA backend?

Additional Context

[Add any additional information here] This is important as we are paying these people to set up this VM for us and later changes will incur additional costs. We plan to train an acoustic model on AWS and export it as a bin to use for decoding on the remote machine. The remote machine only needs the decoding feature of wav2letter.

[abhinavkulkarni]

@nchp89: GPU is not required to run inference - streaming or otherwise.

You can check out the CPU Dockerfile for streaming inference or CPU Dockerfile for non-streaming inference.

[vchagari]

@abhinavkulkarni : Could you able to add CUDA/GPU backend support for Streaming Convnets Streaming inference pipeline?.