Nvidia Audio Effects SDK models

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

I am interested in using the audio noise filter from Nvidia in my app, and I installed the Audio Effects SDK, which provides pre-built trt engines. However, my tensort app fails to deserialize the models with this error:

1: [stdArchiveReader.cpp::nvinfer1::rt::StdArchiveReader::StdArchiveReader::30] Error Code 1: Serialization (Serialization assertion magicTagRead == kMAGIC_TAG failed.Magic tag does not match)

The trtexec.exe app from TensorRT-8.6.1.6 also fails with same error, unless I take the nvinfer.dll file from Audio Effects SDK folder and replace the original in TensorRT-8.6.1.6 folder. However, this clearly isn’t the right way as it causes version mismatch and improper linkage.

So, whoever manages the Audio Effects SDK should rebuild the models using TensorRT-8.6.1.6 and CUDA 12.1, or provide onnx for manual rebuilding. The download page already has 3 versions of the SDK for different GPU architecture, but this isn’t enough due to TensorRT and CUDA versions being different and causing integration problems

Please provide new models

1 Like
2

Hi @fig_tatam ,
Noted your feedback, which will be shared with the team.
Thank you