DeepStream Reference Application - deepstream-audio app¶
Audio application builds on top of the deepstream-app reference application to demonstrate inference of audio data.
deepstream-audio application is open sourced and packaged with DeepStream 5.1 as an alpha product. The AI model used to demonstrate audio inference is an ONNX model. The model file is available within the DeepStream 5.1 package at:
Quick run instructions:
$ cd /opt/nvidia/deepstream/deepstream/sources/apps/sample_apps/deepstream-audio/configs $ deepstream-audio -c ds_audio_sonyc_test_config.txt
ONNX Parser replace instructions (x86 only)¶
The public TensorRT 7.2.2 ONNX parser library fails to parse this model file. The fix for this bug is available in the TensorRT OSS repository. You may build the library and replace the TensorRT 7.2.2 onnxparser library to try the deepstream-audio sample application configs.
Build TRT OSS for onnx parser from
master branch at:
Follow the documentation at TensorRT repository to build TRT from
Commit hash information for the commit in
masterthat was tested:
commit a7e977490a5e143f1df071cf6d6c99becc5ccc88 (HEAD -> master, origin/master, origin/HEAD) Merge: d81f440 60ceef1
Instructions to take backup and copy the newly built onnxparser library:
cp /usr/lib/x86_64-linux-gnu/libnvonnxparser.so.7.2.2 ~/bkup/ sudo cp libnvonnxparser.so.7.2.2 /usr/lib/x86_64-linux-gnu/libnvonnxparser.so.7.2.2
This fix is not required to try out deepstream-audio app on Jetson.