Accent Screech

Accent Preserving Voice Cloning using SV2TTS

Data Preparation

• Run python src/data/data_Create.py
  • Downloads the data into data/raw
  • Creates interim data formats data/interim
  • Creates Processed data data/processed
  • Does other structural setups

Running Models

Run any of the model file directly, each has info regarding how to run it. These files create an accent enocder model that generates accent embeddings. These can be plugged into a syntehsizer such as tacotron to generate accented voice.

• Accent Encoder LSTM Model at src/models/encoder.py
• Accent Encoder 1dConv Model at src/models/encoder_conv.py
• Accent Encoder LSTM Model wiht triplet loss at src/models/encoder_trip.py
• Accent Encoder autoencoder 1DConv Model at src/models/encoder_ae.py
• Accent Encoder autoencoder LSTM Model at src/models/encoder_ae_rec.py
• Accent Encoder Conv VAE Model at src/models/encoder_vae_conv.py

Training metrics are saved in runs directory, can be visualized by pointing tensorboard at the driectory

Models are saved in models_ directory

Project based on the cookiecutter data science project template. #cookiecutterdatascience