Release notes
Current: v2.0
mahkeh-ear v2.0: Kpelle speech to text
CurrentNovember 2026
What’s new
- Listens to Kpelle and writes it down, as well as Koloqua and English.
Known limits
Kpelle is our latest release in its early stage, and fast speech, heavy slang and noisy recordings still cause mistakes.
mahkeh-ear v1.0: Koloqua and English speech to text (first release)
October 2026
What’s new
- Listens to Koloqua or English and writes down what was said, in the language you choose.
- Handles the way Liberians mix Koloqua and English in one sentence.
Benchmark
Held-out recordings it never trained on, measured as word error rate (lower is better).
Known limits
Fast speech, heavy slang and noisy recordings still cause mistakes.
