Addis Scribe now transcribes Sidama, live and from uploads. We compared it with the Ethio-ASR models and Google Gemini on the same held-out natural speech, with the same scorer for every system. Ethio-ASR Sidaama is slightly ahead of us, and this page shows it.
Addis Scribe is within a point of Ethio-ASR Sidaama, the most accurate system here, and has less than half the errors of Google Gemini. It is the only system here that transcribes live.
On 150 WAXAL test clips, streaming in 1.12 s steps, Addis Scribe records 33.6% WER and 9.0% CER. Ethio-ASR Sidaama is slightly ahead at 33.0% WER and 7.3% CER, offline, and Ethio-ASR multilingual 600M records 34.3%.
Read the full accuracy tablesGiven the whole clip, Google Gemini 3 Flash (preview) records 76.0% WER and Gemini Flash (latest) 81.5%.
Read the full accuracy tables150 clips from the WAXAL Sidama test split, natural speech with human transcripts. Every system transcribed the same clips.
WAXAL Sidama, 150 clips, lower is better
WAXAL Sidama, 150 clips, lower is better
| System | Mode | WER | CER |
|---|---|---|---|
| Addis Scribe Sidama | Live, 1.12 s chunks | 33.6% | 9.0% |
| Addis Scribe Sidama | Live, 0.32 s chunks | 34.2% | 9.3% |
| Addis Scribe Sidama | Whole file | 35.0% | 10.2% |
| Ethio-ASR Sidaama | Offline | 33.0% | 7.3% |
| Ethio-ASR multilingual 600M | Offline | 34.3% | 7.7% |
| Google Gemini 3 Flash (preview) | Whole file | 76.0% | 21.4% |
| Google Gemini Flash (latest) | Whole file | 81.5% | 36.6% |
Bold marks the best value in each column. The charts cut the Gemini bars; their values are in the table. Addis Scribe has a higher character error rate than the Ethio-ASR models here, while its word error rate is within a point of the best.
Addis Scribe streams live audio. The other systems were given the whole recording, so they are scored for accuracy only.
| System | Live audio | How it was run |
|---|---|---|
| Addis Scribe Sidama | Native streaming | Streams in 0.32 s or 1.12 s chunks through the Addis Scribe API, so text appears while people speak. |
| Ethio-ASR Sidaama | Offline only | Needs the whole recording before it can answer. Scored for accuracy. |
| Ethio-ASR multilingual 600M | Offline only | Needs the whole recording before it can answer. Scored for accuracy. |
| Google Gemini 3 Flash (preview) | Offline only | Given the whole clip with a transcription prompt. No live mode, so scored for accuracy. |
| Google Gemini Flash (latest) | Offline only | Given the whole clip with a transcription prompt. No live mode, so scored for accuracy. |
The test set is public. None of these test clips were used to train Addis Scribe.
| Test set | Where to find it | Clips | What it measures |
|---|---|---|---|
| WAXAL | google/WaxalNLP | Sidama test split, 150 of its 3,561 clips | Natural speech with human transcripts |
Every system on a set is scored with the same scorer and text normaliser, against the same human transcripts. WER is word error rate and CER is character error rate, both lower is better.
Add language=sid to a Scribe upload, live session or long-file job. Uploads, live streaming, word timestamps, captions and speaker labels all work in Sidama, at the same per-character price as Amharic. Sidama text comes back in lowercase without punctuation; an apostrophe marks the glottal stop.