🎯 English Speech2IPA Leaderboard

Developed By: Koel Labs

Evaluation

We use two standard metrics:

• PER (Phoneme Error Rate): The Levenshtein distance calculated between phoneme sequences of the predicted and actual transcriptions.
• FER (Feature Error Rate): The edit distance between the predicted and actual phoneme sequences, weighted by the phonetic features from panphon.

Models are evaluated on a variety of English speech: native, non-native, and impaired. Read more about evaluations or how to build your own leaderboards on our blog.

Compute

This leaderboard uses the free basic plan (16GB RAM, 2vCPUs) to allow for reproducability. The evaluation may take several hours to complete. Please be patient and do not submit the same model multiple times.

Contributing, Questions, and Feedback

Please read the README.md for more information on how to contribute, ask questions, or provide feedback.