Enter the SIDEKIT

• How to manage the data: IdMap, Ndx, Key, Scores and StatServer
  • IdMap
  • Ndx
  • Keys
  • Scores
  • StatServer
• Parallel computation in SIDEKIT
  • Multiprocessing
  • MPI
• Acoustic parametrization
  • 1. Save the features in HDF5 format
  • 2. The FeaturesExtractor object
  • 3. The FeaturesServer object
• Train a Universal Background Model
  • 1. Training using EM split
  • 2. Training using simple EM with fixed number of distributions
  • 3. Training using EM split on several nodes
  • 4 Full covariance UBM
• Train an i-vector extractor
  • 1. Get to know the algorithm with total_variability_raw
  • 2. Using a single process on one machine
  • 3. Using multiple process on one machine with Python MultiProcessing
  • 4. Using multiple process on multiple nodes with MPI
• Extract your I-Vectors
  • 1. Extract i-vectors in a single process
  • 2. Extract i-vectors on multiple process on a single node
  • 3. Extract i-vectors on multiple nodes
• Train a deep neural network with Theano and SIDEKIT
  • Requirement
  • Train your Feed-Forward DNN
• Bottleneck features extraction
• Phonetically aware Neural Network for speaker recognition