audonnx 0.6.0

Deploy models in ONNX format

audonnx deploys machine learning models stored in ONNX format.

Machine learning models can be trained in a variety of frameworks, e.g. scikit-learn, TensorFlow, Torch. To be independent of the training framework and its version models can be exported in ONNX format, which enables you to deploy and combine them easily.

audonnx allows you to name inputs and outputs of your model, and automatically loads the correct feature extraction from a YAML file stored with your model.

Have a look at the installation and usage instructions.

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Version 0.6.0 (2022-06-23)

• Added: audonnx.Function

• Added: audonnx.Model.labels()

• Added: arguments concat, outputs, squeeze to audonnx.Model.__call__()

• Added: tests on Windows

• Added: audonnx.testing module

• Changed: optionally init audonnx.Model from proto object instead of ONNX file

• Changed: dynamic axis can be specified as None in ONNX graph

• Changed: support output nodes where last dimension is dynamic

• Deprecated: argument output_names of audonnx.Model.__call__()

Version 0.5.2 (2022-04-01)

• Fixed: always replace dynamic axis names with -1 in input and output shapes of model nodes

Version 0.5.1 (2022-03-29)

• Added: argument auto_install to audonnx.load()

Version 0.5.0 (2022-02-09)

• Added: argument device

• Changed: use CPU by default

• Changed: require onnxruntime>=1.8.0

• Removed: audonnx.Model.forward() audonnx.Model.labels, audonnx.Model.predict(), audonnx.Model.transform

Version 0.4.3 (2022-01-10)

• Fixed: publication of docs failed

Version 0.4.2 (2022-01-10)

• Fixed: publication of docs failed

Version 0.4.1 (2022-01-10)

• Fixed: author email address in Python package metadata

Version 0.4.0 (2022-01-10)

• Added: first public release

• Changed: switch to MIT license

• Changed: move repo to Github

• Fixed: remove audsp from docstring example as we no longer depend on it

Version 0.3.3 (2021-12-30)

• Changed: use Python 3.8 as default

Version 0.3.2 (2021-11-01)

• Changed: use audobject >=0.6.1

Version 0.3.1 (2021-10-05)

• Fixed: audonnx.load() try to load model from ONNX if YAML does not exist

Version 0.3.0 (2021-10-01)

• Changed: audobject >=0.5.0

• Changed: force .yaml extension when model is saved

• Fixed: if possible load model from .yaml in audonnx.load()

Version 0.2.2 (2021-09-23)

• Fixed: link to ONNX runtime CUDA mapping table

Version 0.2.1 (2021-09-15)

• Fixed: loading of old models that contain a model.yaml file

Version 0.2.0 (2021-07-20)

• Added: audonnx.InputNode, audonnx.Model.__call__(), audonnx.Model.inputs, audonnx.Model.outputs, audonnx.OutputNode

• Changed: reshape input to expected shape

• Changed: do not depend on existing models in tests and documentation

• Changed: support multiple input nodes

• Changed: make audonnx.Model serializable

• Deprecated: audonnx.Model.forward() audonnx.Model.labels, audonnx.Model.predict(), audonnx.Model.transform

• Removed: audonnx.Model.input_node, audonnx.Model.input_shape, audonnx.Model.input_type, audonnx.Model.output_nodes, audonnx.Model.output_shape, audonnx.Model.output_type,

Version 0.1.1 (2021-03-31)

• Changed: update documentation how to select specific GPU device

Version 0.1.0 (2021-03-25)

• Added: initial release