# YAMNet third-party notices

Una Corda's optional piano-activity detector is a modified, ONNX-converted form
of Google's YAMNet v1 model and TensorFlow Models YAMNet source.

- Source: <https://github.com/tensorflow/models/tree/4d7bdd8c170ee90850f2f9ccef0f6d19b817de35/research/audioset/yamnet>
- Official model version: <https://www.kaggle.com/models/google/yamnet/tensorFlow2/yamnet/1>
- TF Hub alias: <https://tfhub.dev/google/yamnet/1>
- Pinned repository license notices: Copyright 2022 Google LLC; Copyright
  2016, The Authors.
- YAMNet source headers: Copyright 2019 The TensorFlow Authors.
- License: Apache License 2.0; the complete text is in `APACHE-2.0.txt`.

The official model-version metadata explicitly labels the model `Apache 2.0`.
Una Corda converted the official TensorFlow SavedModel to ONNX opset 17, then
modified 28 learned Conv/classifier initializers to fp16 storage with explicit
fp32 casts. The official waveform frontend, model I/O, biases, activations, and
compute remain fp32. The converted file is named with its own SHA-256 prefix.

No NOTICE file was found at the pinned TensorFlow Models repository root. If a
future pinned source or artifact supplies NOTICE content, preserve it with any
redistribution. Google, TensorFlow, YAMNet, AudioSet, and related marks belong
to their respective owners. This project is not endorsed by Google.

AudioSet is a YouTube-derived, weakly labeled training corpus. Its provenance
and dataset limitations are recorded separately from the model copyright
license and do not establish fitness for any household deployment.