v2021.1.1

This release includes minor changes involving updating URLs and language within the README and in the repo to reflect the renaming of this organization from neuralaudio to hearbenchmark.

What's Changed

• Update Organization by @jorshi in #43
• v2021.1.1 by @jorshi in #44

Full Changelog: v2021.1.0...v2021.1.1

v2021.1.0

HEAR 2021 Competition Version

• Minor fix, unpins transformers version

v2021.0.5

• wav2vec2 baseline uses facebook/wav2vec2-large-100k-voxpopuli model by default and has different hop sizes for scene and timestamp embedding.
• torchcrepe baseline has different hop sizes for scene and timestamp embedding.

v2021.0.3

• torch.fft.rfft fix for pytorch 1.7
• tensorflow model now loads weights so it matches the pytorch implementation
• Keras implementation of baseline model
• hearbaseline.wav2vec2 model

v2021.0.2

• Adds a TensorFlow version of the naive baseline model
• Updates the timestamps returned from get_timestamp_embeddings to have a shape (n_sounds, n_timestamps)

v2021.0.1

HEAR 2021 naïve baseline model

• PyTorch implementation of naive baseline: mel-spectrogram with random projection