v0.5.0

* Release `AddBackgroundNoise` and `ApplyImpulseResponse`

* Implement `Shift`
* Remove support for 1-dimensional and 2-dimensional audio tensors. Only 3-dimensional audio
 tensors are supported now.
* Make `sample_rate` optional. Allow specifying `sample_rate` in `__init__` instead of `forward`. This means torchaudio transforms can be used in `Compose` now.
* Fix a bug where one could not use the `parameters` method of the `nn.Module` subclass