Swahili speech-to-text API: 34% word error, measured

Measured accuracy

Model	Word error rate	Character error rate	Clips with no errors	Empty transcripts	Speed
Whisper Large v3 `openai/whisper-large-v3`	34%	8.0%	0 of 30	0 of 30	0.052× real time

Fair. Gets the content across, but needs a human pass before it is published or quoted. The clips are clean, single-speaker read speech, so treat this as a best case for your own recordings. How we measured, and all 83 languages.

What a transcript looks like

The clip with the median error rate from the test, reference text against what the API returned.

Spoken	Mtiririko wa Trafiki ni utafiti wa mwondoko wa madereva binafsi na magari kati ya sehemu mbili na mwingiliano wanaofanya na wenzao.
Whisper Large v3	Ntiriko wa trafiki ni utafiti wa mundoko wa mederewa binazi na magari kati ya sehemu mbili na mingiliano wanawafanya na mzau.

Language detection

All 10 clips we sent without a language field were identified as Swahili. Auto-detection is safe here; sending language=sw still skips the detection step.

Quickstart

The request is the OpenAI transcription request with a different base URL. language is optional.

cURL

curl https://api.inferenceapis.com/v1/audio/transcriptions \
  -H "Authorization: Bearer $INFERENCE_API_KEY" \
  -F "model=openai/whisper-large-v3" \
  -F "language=sw" \
  -F "file=@recording.mp3" \
  -F "response_format=verbose_json"

Python · openai SDK

import os
from openai import OpenAI

client = OpenAI(base_url="https://api.inferenceapis.com/v1", api_key=os.environ["INFERENCE_API_KEY"])
with open("recording.mp3", "rb") as f:
    r = client.audio.transcriptions.create(model="openai/whisper-large-v3", file=f, language="sw", response_format="verbose_json")
print(r.text)

Subtitles and timestamps

response_format=srt or vtt returns a subtitle file with cues of up to two lines, broken at sentence ends and pauses. verbose_json with timestamp_granularities[]=word returns a start and end time for every word if you want to build your own.

Cost

$0.002 per minute of audio on either model: $0.12 for an hour, $12 for a hundred hours. Duration is measured from the file you upload. There is no daily cap on minutes and no subscription; enough to run your own Swahili test.

FAQ

How accurate is Whisper for Swahili?

On 30 clean read-speech clips it made 34% word errors, about one word in 3. Gets the content across, but needs a human pass before it is published or quoted. Real recordings with noise, accents or several speakers will score worse.

Does language auto-detection work for Swahili?

In our test, yes: all 10 clips sent without a language were identified as Swahili. Sending language=sw still saves the detection step on every request.

Can it translate Swahili speech to English?

The API returns Swahili text. For English, transcribe first and translate the text with a chat model on the same key.

Related languages: Shona (over 100%) · Somali (93%) · Lingala (66%)