config.yml

# The MIT License (MIT)
#
# Copyright (c) 2018 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.
#
# AUTHOR
# Hervé Bredin - http://herve.niderb.fr

# train the network using triplet loss
# see pyannote.audio.embedding.approaches for more details
approach:
  name: TripletLoss
  params:
    metric: cosine    # embeddings are optimized for cosine metric
    clamp: positive   # triplet loss variant
    margin: 0.2       # triplet loss margin
    duration: 2       # sequence duration
    sampling: all     # triplet sampling strategy
    per_fold: 40      # number of speakers per fold
    per_label: 3      # number of sequences per speaker

# use precomputed features (see feature extraction tutorial)
feature_extraction:
  name: Precomputed
  params:
     root_dir: tutorials/feature-extraction

# use the TristouNet architecture.
# see pyannote.audio.embedding.models for more details
architecture:
  name: TristouNet
  params:
    rnn: LSTM
    recurrent: [16]
    bidirectional: True
    pooling: sum
    linear: [16, 16]

# use cyclic learning rate scheduler
scheduler:
  name: CyclicScheduler
  params:
      learning_rate: auto