The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Reverts to the weights from RNNoise, which work better for voice detection.
- Adds a more user-friendly API, based on the
dasp
crate. This adds a dependency ondasp
, which can be removed by disabling the (on-by-default) "dasp" feature.
- Publically exposes some training-related functions, even when the
train
feature is disabled.
- New
train
feature allows for generating training features. - Documentation for training new available in
train/README.md
. RnnModel::from_static_bytes
allows for loading models without allocation.
RnnModel::from_read
replaced byRnnModel::from_bytes
. (Breaking change!)