asr_eval ⚗️

Evaluation and building components for Automatic Speech Recognition in Python.

• Preprint paper
• Full documentation

Installation

Run pip install asr_eval[all] (see installation guide for more details).

Features

🧪 Evaluation features (see overview, guide and reference)

• Extended annotation syntax via multi-reference and wildcard blocks
• Aligned multiple model comparision with error highlighting

⚙️ Benchmarking features (see guide and reference)

• A set of baseline pipelines and datasets (in progress)
• Model benchmarking in scale with a web dashboard
• Components to build custom pipelines

➰ Streaming features (see overview, guide and reference)

• A base class and buffers for streaming models
• Streaming evaluation diagrams

AI disclaimer

Currently, LLM coding is used only for few non-critical components (streaming dashboard, CI pipeline). We also plan to employ LLM to generate regression tests, given docstring specifications.

Contributing

This project is in active development stage. Feel free to contrubite:

• Open an Issue on Github
• Open a Pull request (Contributing guide in progress)
• Write in Telegram chat and tag @olegs9