Speech-to-Text for the Ukrainian language based on Silero

This is a repository with demonstration code that uses the Silero Model for Ukrainian in the task of Speech-to-Text recognition.

Quality:

Common Voice 7 test set with 4300+ samples:

WER: 0.2318 (id est - quality is 76.82%)
CER: 0.0624

How to run

Linux

Install dependencies and enter the python environment:

pipenv install
pipenv shell

Run the demos:

python test_official_demo.py
python own_recodings_demo.py

Windows

1. Install Anaconda (from: https://www.anaconda.com/products/individual#windows)

2. Run Anaconda Powershell Prompt

3. Install PyTorch and other libraries:

   conda install pytorch torchvision torchaudio cpuonly -c pytorch
   

4. Install other libraries:

   pip install SoundFile
   

Run the demos:

cd C:\path\to\project

python .\test_official_demo.py
python .\own_recodings_demo.py

It was tested on Windows x64 only.