Ukrainian ASR from Respeecher

Training code for a Ukrainian ASR system is based on the Data2Vec model, which was published by Respeecher. Respeecher provides Hollywood-quality speech-to-speech and text-to-speech AI voices to businesses and various types of content creators.

Setup Working Environment

Pre-requirements

• conda 4.12.0 (later versions may also work) - Installation
• (Optional) CUDA Version: 11.4; Driver Version: 470.129.06 - Installation

Setup environment

conda env create -f environment.yaml

Train

CUDA_VISIBLE_DEVICES="{gpu}" python torch_train.py

Results

Best model can be found in logdirs/torch_asr_on_ukrainian_data2vec_cosinev3/best_model

Metric JSON can be found in logdirs/torch_asr_on_ukrainian_data2vec_cosinev3/metric.json