| # Attribution and license scope |
|
|
| This is a mixed-license-scope repository. Fluid Inference intends the included |
| CC-BY-4.0 license text to cover only the following exact Community-1-derived |
| artifacts and their uncompiled counterparts listed in `provenance.json`: |
|
|
| - `Segmentation.mlmodelc` |
| - `FBank.mlmodelc` |
| - `Embedding.mlmodelc` |
| - `PLDA.mlmodelc` |
| - `PldaRho.mlmodelc` |
| - `plda-parameters.json` |
| - `xvector-transform.json` |
|
|
| The underlying Community-1 pipeline is published by pyannote under CC-BY-4.0: |
| https://huggingface.co/pyannote/speaker-diarization-community-1 |
|
|
| The PLDA parameters originate from Brno University of Technology / BUT |
| Speech@FIT. The rights holder explicitly licenses `plda.npz` and |
| `xvec_transform.npz` under CC-BY-4.0, including commercial use: |
| https://huggingface.co/BUT-FIT/diarizen-wavlm-large-s80-md/blob/6285693ddd5b38e8229acb93f864f3d04a82bee1/plda/LICENSE |
|
|
| Fluid Inference converted the PyTorch components to Core ML, introduced fixed |
| and enumerated input shapes, applied mixed-precision storage where recorded in |
| the model metadata, separated the FBank frontend from the embedding backend, |
| and compiled packages for Apple platforms. |
|
|
| Appropriate attribution should identify pyannote, WeSpeaker, BUT Speech@FIT, |
| and Fluid Inference, retain the citations in README.md, link CC-BY-4.0, and |
| indicate that the files are modified Core ML conversions. |
|
|
| ## Legacy artifacts |
|
|
| `pyannote_segmentation.mlmodelc`, `wespeaker.mlmodelc`, |
| `wespeaker_v2.mlmodelc`, `wespeaker_int8.mlmodelc`, and their packages predate |
| the Community-1 artifact set and are not covered by the Community-1 scope |
| above. They are Core ML conversions of the pyannote/speaker-diarization-3.1 |
| components, under their own upstream licenses: |
|
|
| | Artifact | Upstream checkpoint | License | |
| |---|---|---| |
| | `pyannote_segmentation` | [`pyannote/segmentation-3.0`](https://huggingface.co/pyannote/segmentation-3.0) @ `e66f3d3b` (Copyright (c) 2023 CNRS) | MIT | |
| | `wespeaker` (fp32) | [`pyannote/wespeaker-voxceleb-resnet34-LM`](https://huggingface.co/pyannote/wespeaker-voxceleb-resnet34-LM) @ `837717dd`, wrapping WeSpeaker's VoxCeleb ResNet34-LM | CC-BY-4.0 | |
| | `wespeaker_v2`, `wespeaker_int8` (byte-identical; 8-bit palettized weights) | same as `wespeaker` | CC-BY-4.0 | |
|
|
| The original conversion scripts were not preserved, so this was confirmed by |
| output parity on AMI ES2004a against the PyTorch checkpoints above (pyannote.audio |
| 3.3.2, CPU): segmentation frame-level argmax agreement 99.98%; speaker |
| embeddings cosine 1.0000 (`wespeaker`) and >= 0.997 (`wespeaker_v2`/`wespeaker_int8`). |
|
|
| Fluid Inference converted these checkpoints to Core ML with fixed input shapes, |
| fused the fbank frontend into the embedding model, and palettized the |
| `wespeaker_v2`/`wespeaker_int8` weights to 8 bits. Attribution for the |
| embedding models should identify WeSpeaker, pyannote, and Fluid Inference, |
| link CC-BY-4.0, and indicate that the files are modified Core ML conversions. |
| The segmentation model's MIT notice is reproduced below. |
|
|
| ``` |
| MIT License |
| |
| Copyright (c) 2023 CNRS |
| |
| Permission is hereby granted, free of charge, to any person obtaining a copy |
| of this software and associated documentation files (the "Software"), to deal |
| in the Software without restriction, including without limitation the rights |
| to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
| copies of the Software, and to permit persons to whom the Software is |
| furnished to do so, subject to the following conditions: |
| |
| The above copyright notice and this permission notice shall be included in all |
| copies or substantial portions of the Software. |
| |
| THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| SOFTWARE. |
| ``` |
|
|
| Fluid Inference can grant rights only to the extent it is authorized to do so. |
| This notice does not restrict or replace rights granted directly by upstream |
| licensors. |
|
|
