/stt-benchmarking

a place to discuss and create an independent speech-to-text benchmarking tool

Primary LanguagePython

stt-benchmarking

a place to discuss and create an independent speech-to-text benchmarking tool.

Design principles

  • Benchmarking metrics will be measured in relation to a reference transcript and audio prepared by the client.
  • The tool should be language-agnostic: the client will create the reference in the language they require and test the tools.
  • The tool should require as little configuration as possible and depend on test data format as much as possible for deciding which evaluation metrics to compute. (R5)

Installation

  • python3 -m venv venv
  • pip install -r requirements.txt
  • python -m spacy download en

TextAV STT-Benchmarking GDoc