Addis Voice 2 on Amharic
Addis Voice 2 against Azure, Gemini, OpenAI, ElevenLabs and Meta MMS on 100 Amharic prompts from three open datasets, with audio, judge transcripts and confidence intervals.
Read the benchmarkBenchmarksHow our models measure up,
How our models measure up,
with the workings shown.
We publish the datasets, the pinned revisions, the scoring rules, the audio, the artifact hashes and the limitations for every run, including the results where another system comes out ahead. If you want to check us, everything you need is on these pages or in the archive.
Published runs
How we publish
What every run on this page includes
Fixed inputs
Public test-split prompts pinned to exact dataset revisions and drawn once with a fixed seed, so the prompt set can be rebuilt independently and hash-checked against ours.
Full results
Confidence intervals on every error rate, per-slice breakdowns, latency, and the rows where a competitor leads, highlighted for them rather than left out.
Named conditions
Systems without documented support for the language are labelled as unsupported-language conditions, and the model and voice used for every system is published.
Every run is published as an open dataset with the prompts, the generated audio, the judge transcripts, the scorer and a checksum for every file. The Amharic run is at addisai/amharic-tts-benchmark.