Upload 10 files

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2022 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.

README.md

@@ -1,3 +1,121 @@
----
-license: mit
----
+---
+tags:
+- pyannote
+- pyannote-audio
+- pyannote-audio-model
+- audio
+- voice
+- speech
+- speaker
+- speaker-recognition
+- speaker-verification
+- speaker-identification
+- speaker-embedding
+datasets:
+- voxceleb
+license: mit
+inference: false
+extra_gated_prompt: "The collected information will help acquire a better knowledge of pyannote.audio userbase and help its maintainers apply for grants to improve it further. If you are an academic researcher, please cite the relevant papers in your own publications using the model. If you work for a company, please consider contributing back to pyannote.audio development (e.g. through unrestricted gifts). We also provide scientific consulting services around speaker diarization and machine listening."
+extra_gated_fields:
+  Company/university: text
+  Website: text
+  I plan to use this model for (task, type of audio data, etc): text
+---
+
+Using this open-source model in production?  
+Consider switching to [pyannoteAI](https://www.pyannote.ai) for better and faster options.
+
+# 🎹 Speaker embedding
+
+Relies on pyannote.audio 2.1: see [installation instructions](https://github.com/pyannote/pyannote-audio/).
+
+This model is based on the [canonical x-vector TDNN-based architecture](https://ieeexplore.ieee.org/abstract/document/8461375), but with filter banks replaced with [trainable SincNet features](https://ieeexplore.ieee.org/document/8639585). See [`XVectorSincNet`](https://github.com/pyannote/pyannote-audio/blob/3c988c028dc505c64fe776720372f6fe816b585a/pyannote/audio/models/embedding/xvector.py#L104-L169) architecture for implementation details.
+
+
+## Basic usage
+
+```python
+# 1. visit hf.co/pyannote/embedding and accept user conditions
+# 2. visit hf.co/settings/tokens to create an access token
+# 3. instantiate pretrained model
+from pyannote.audio import Model
+model = Model.from_pretrained("pyannote/embedding", 
+                              use_auth_token="ACCESS_TOKEN_GOES_HERE")
+```
+
+```python
+from pyannote.audio import Inference
+inference = Inference(model, window="whole")
+embedding1 = inference("speaker1.wav")
+embedding2 = inference("speaker2.wav")
+# `embeddingX` is (1 x D) numpy array extracted from the file as a whole.
+
+from scipy.spatial.distance import cdist
+distance = cdist(embedding1, embedding2, metric="cosine")[0,0]
+# `distance` is a `float` describing how dissimilar speakers 1 and 2 are.
+```
+
+Using cosine distance directly, this model reaches 2.8% equal error rate (EER) on VoxCeleb 1 test set.   
+This is without voice activity detection (VAD) nor probabilistic linear discriminant analysis (PLDA).
+Expect even better results when adding one of those.
+
+## Advanced usage
+
+### Running on GPU
+
+```python
+import torch
+inference.to(torch.device("cuda"))
+embedding = inference("audio.wav")
+```
+
+### Extract embedding from an excerpt
+
+```python
+from pyannote.audio import Inference
+from pyannote.core import Segment
+inference = Inference(model, window="whole")
+excerpt = Segment(13.37, 19.81)
+embedding = inference.crop("audio.wav", excerpt)
+# `embedding` is (1 x D) numpy array extracted from the file excerpt.
+```
+
+### Extract embeddings using a sliding window
+
+```python
+from pyannote.audio import Inference
+inference = Inference(model, window="sliding",
+                      duration=3.0, step=1.0)
+embeddings = inference("audio.wav")
+# `embeddings` is a (N x D) pyannote.core.SlidingWindowFeature
+# `embeddings[i]` is the embedding of the ith position of the 
+# sliding window, i.e. from [i * step, i * step + duration].
+```
+
+
+## Citation
+
+```bibtex
+@inproceedings{Bredin2020,
+  Title = {{pyannote.audio: neural building blocks for speaker diarization}},
+  Author = {{Bredin}, Herv{\'e} and {Yin}, Ruiqing and {Coria}, Juan Manuel and {Gelly}, Gregory and {Korshunov}, Pavel and {Lavechin}, Marvin and {Fustes}, Diego and {Titeux}, Hadrien and {Bouaziz}, Wassim and {Gill}, Marie-Philippe},
+  Booktitle = {ICASSP 2020, IEEE International Conference on Acoustics, Speech, and Signal Processing},
+  Address = {Barcelona, Spain},
+  Month = {May},
+  Year = {2020},
+}
+```
+
+```bibtex
+@inproceedings{Coria2020,
+    author="Coria, Juan M. and Bredin, Herv{\'e} and Ghannay, Sahar and Rosset, Sophie",
+    editor="Espinosa-Anke, Luis and Mart{\'i}n-Vide, Carlos and Spasi{\'{c}}, Irena",
+    title="{A Comparison of Metric Learning Loss Functions for End-To-End Speaker Verification}",
+    booktitle="Statistical Language and Speech Processing",
+    year="2020",
+    publisher="Springer International Publishing",
+    pages="137--148",
+    isbn="978-3-030-59430-5"
+}
+```
+

config.yaml

@@ -0,0 +1,85 @@
+protocol: VoxCeleb.SpeakerVerification.VoxCeleb_X
+patience: 5
+task:
+  _target_: pyannote.audio.tasks.SupervisedRepresentationLearningWithArcFace
+  min_duration: 2
+  duration: 5.0
+  num_classes_per_batch: 64
+  num_chunks_per_class: 4
+  margin: 10.0
+  scale: 50.0
+  num_workers: 20
+  pin_memory: false
+model:
+  _target_: pyannote.audio.models.embedding.XVectorSincNet
+optimizer:
+  _target_: torch.optim.Adam
+  lr: 0.001
+  betas:
+  - 0.9
+  - 0.999
+  eps: 1.0e-08
+  weight_decay: 0
+  amsgrad: false
+trainer:
+  _target_: pytorch_lightning.Trainer
+  accelerator: null
+  accumulate_grad_batches: 1
+  amp_backend: native
+  amp_level: O2
+  auto_lr_find: false
+  auto_scale_batch_size: false
+  auto_select_gpus: true
+  benchmark: false
+  check_val_every_n_epoch: 1
+  checkpoint_callback: true
+  deterministic: false
+  fast_dev_run: false
+  flush_logs_every_n_steps: 100
+  gpus: 1
+  gradient_clip_val: 0
+  limit_test_batches: 1.0
+  limit_train_batches: 1.0
+  limit_val_batches: 1.0
+  log_every_n_steps: 50
+  log_gpu_memory: null
+  max_epochs: 1000
+  max_steps: null
+  min_epochs: 1
+  min_steps: null
+  num_nodes: 1
+  num_processes: 1
+  num_sanity_val_steps: 2
+  overfit_batches: 0.0
+  precision: 32
+  prepare_data_per_node: true
+  process_position: 0
+  profiler: null
+  progress_bar_refresh_rate: 1
+  reload_dataloaders_every_epoch: false
+  replace_sampler_ddp: true
+  sync_batchnorm: false
+  terminate_on_nan: false
+  tpu_cores: null
+  track_grad_norm: -1
+  truncated_bptt_steps: null
+  val_check_interval: 1.0
+  weights_save_path: null
+  weights_summary: top
+augmentation:
+  transform: Compose
+  params:
+    shuffle: false
+    transforms:
+    - transform: AddBackgroundNoise
+      params:
+        background_paths: /gpfswork/rech/eie/commun/data/background/musan
+        min_snr_in_db: 5.0
+        max_snr_in_db: 15.0
+        mode: per_example
+        p: 0.9
+    - transform: ApplyImpulseResponse
+      params:
+        ir_paths: /gpfswork/rech/eie/commun/data/rir
+        mode: per_example
+        p: 0.5

gitattributes

@@ -0,0 +1,16 @@
+*.bin.* filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tar.gz filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text

hparams.yaml

@@ -0,0 +1,6 @@
+sample_rate: 16000
+num_channels: 1
+sincnet:
+  stride: 10
+  sample_rate: 16000
+dimension: 512

hydra.yaml

@@ -0,0 +1,139 @@
+hydra:
+  run:
+    dir: ${protocol}/${task._target_}/${now:%Y-%m-%d}/${now:%H-%M-%S}
+  sweep:
+    dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S}/${protocol}/${task._target_}
+    subdir: ${hydra.job.num}
+  hydra_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][HYDRA] %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+    root:
+      level: INFO
+      handlers:
+      - console
+    loggers:
+      logging_example:
+        level: DEBUG
+    disable_existing_loggers: false
+  job_logging:
+    version: 1
+    formatters:
+      simple:
+        format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s'
+    handlers:
+      console:
+        class: logging.StreamHandler
+        formatter: simple
+        stream: ext://sys.stdout
+      file:
+        class: logging.FileHandler
+        formatter: simple
+        filename: ${hydra.job.name}.log
+    root:
+      level: INFO
+      handlers:
+      - console
+      - file
+    disable_existing_loggers: false
+  sweeper:
+    _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper
+    max_batch_size: null
+  launcher:
+    _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher
+  help:
+    app_name: pyannote-audio-train
+    header: == ${hydra.help.app_name} ==
+    footer: 'Powered by Hydra (https://hydra.cc)
+
+      Use --hydra-help to view Hydra specific help'
+    template: "${hydra.help.header}\n\npyannote-audio-train protocol={protocol_name}\
+      \ task={task} model={model}\n\n{task} can be any of the following:\n* vad (default)\
+      \ = voice activity detection\n* scd = speaker change detection\n* osd = overlapped\
+      \ speech detection\n* xseg = multi-task segmentation\n\n{model} can be any of\
+      \ the following:\n* debug (default) = simple segmentation model for debugging\
+      \ purposes\n\n{optimizer} can be any of the following\n* adam (default) = Adam\
+      \ optimizer\n\n{trainer} can be any of the following\n* fast_dev_run for debugging\n\
+      * default (default) for training the model\n\nOptions\n=======\n\nHere, we describe\
+      \ the most common options: use \"--cfg job\" option to get a complete list.\n\
+      \n* task.duration: audio chunk duration (in seconds)\n* task.batch_size: number\
+      \ of audio chunks per batch\n* task.num_workers: number of workers used for\
+      \ generating training chunks\n\n* optimizer.lr: learning rate\n* trainer.auto_lr_find:\
+      \ use pytorch-lightning AutoLR\n\nHyper-parameter optimization\n============================\n\
+      \nBecause it is powered by Hydra (https://hydra.cc), one can run grid search\
+      \ using the --multirun option.\n\nFor instance, the following command will run\
+      \ the same job three times, with three different learning rates:\n  pyannote-audio-train\
+      \ --multirun protocol={protocol_name} task={task} optimizer.lr=1e-3,1e-2,1e-1\n\
+      \nEven better, one can use Ax (https://ax.dev) sweeper to optimize learning\
+      \ rate directly:\n  pyannote-audio-train --multirun hydra/sweeper=ax protocol={protocol_name}\
+      \ task={task} optimizer.lr=\"interval(1e-3, 1e-1)\"\n\nSee https://hydra.cc/docs/plugins/ax_sweeper\
+      \ for more details.\n\nUser-defined task or model\n==========================\n\
+      \n1. define your_package.YourTask (or your_package.YourModel) class\n2. create\
+      \ file /path/to/your_config/task/your_task.yaml  (or /path/to/your_config/model/your_model.yaml)\n\
+      \   # @package _group_\n   _target_: your_package.YourTask  # or YourModel\n\
+      \   param1: value1\n   param2: value2\n3. call pyannote-audio-train --config-dir\
+      \ /path/to/your_config task=your_task task.param1=modified_value1 model=your_model\
+      \ ...\n\n${hydra.help.footer}"
+  hydra_help:
+    hydra_help: ???
+    template: 'Hydra (${hydra.runtime.version})
+
+      See https://hydra.cc for more info.
+
+
+      == Flags ==
+
+      $FLAGS_HELP
+
+
+      == Configuration groups ==
+
+      Compose your configuration from those groups (For example, append hydra/job_logging=disabled
+      to command line)
+
+
+      $HYDRA_CONFIG_GROUPS
+
+
+      Use ''--cfg hydra'' to Show the Hydra config.
+
+      '
+  output_subdir: ''
+  overrides:
+    hydra: []
+    task:
+    - protocol=VoxCeleb.SpeakerVerification.VoxCeleb_X
+    - task=SpeakerEmbedding
+    - task.num_workers=20
+    - task.min_duration=2
+    - task.duration=5.
+    - task.num_classes_per_batch=64
+    - task.num_chunks_per_class=4
+    - task.margin=10.0
+    - task.scale=50.
+    - model=XVectorSincNet
+    - trainer.gpus=1
+    - +augmentation=background_then_reverb
+  job:
+    name: train
+    override_dirname: +augmentation=background_then_reverb,model=XVectorSincNet,protocol=VoxCeleb.SpeakerVerification.VoxCeleb_X,task.duration=5.,task.margin=10.0,task.min_duration=2,task.num_chunks_per_class=4,task.num_classes_per_batch=64,task.num_workers=20,task.scale=50.,task=SpeakerEmbedding,trainer.gpus=1
+    id: ???
+    num: ???
+    config_name: config
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.0.4
+    cwd: /gpfsdswork/projects/rech/eie/uno46kl/xvectors/debug
+  verbose: false

overrides.yaml

@@ -0,0 +1,12 @@
+- protocol=VoxCeleb.SpeakerVerification.VoxCeleb_X
+- task=SpeakerEmbedding
+- task.num_workers=20
+- task.min_duration=2
+- task.duration=5.
+- task.num_classes_per_batch=64
+- task.num_chunks_per_class=4
+- task.margin=10.0
+- task.scale=50.
+- model=XVectorSincNet
+- trainer.gpus=1
+- +augmentation=background_then_reverb

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4bcec986de13da7af7ac88736572692359950df63669989c4f78b294934c9089
+size 96383626

tfevents.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3319218e36d416c5400ffbc592acc2e1ab520a187d586be86db7eef30fb65616
+size 5669685