Skip to main content
Khulasa

Benchmark changelog

These figures describe those configurations, on that audio, on that date. Vendors update their models, and the same run repeated later may not reproduce them.

  1. First publication
    Dataset
    UN ESCWA meeting audio, code-switched
    Normaliser
    N1legacy
    Method
    2,000 draws at 95% confidence

    Swipe the table to see every column.

    Models and API versions each system was run at in this benchmark run
    SystemModel or APIVersionNotes
    Khulasa — self-hostedWhisper large-v3Not recorded for this runSelf-hostedPublished reference configuration
    Speechmaticsstandard Arabic/EnglishSpeechmatics Batch APIstandardNot recorded for this run
    Speechmatics melia-1code-switch modelSpeechmatics Batch APImelia-1Not recorded for this run
    Speechmaticsenhanced Arabic/EnglishSpeechmatics Batch APIenhancedNot recorded for this run
    A leading Arabic cloud vendorcode-switch modelWithheld, as the vendor's name is

Back to the full benchmark

Cookies on Khulasa

We use strictly necessary cookies to run the site. With your permission we also use Google Analytics to understand how the site is used. You can change your choice at any time from "Cookie settings" in the footer. Privacy Policy