Speech to text
Make every word usable.
Turn Amharic and Afaan Oromo recordings into text for search, notes, and accessible digital experiences.
Demo by requestFrom Ethiopia. For the way we speak.
Language AI,
built for here.
Turn spoken words and document images into text, and your knowledge into useful answers.
Speech, Amharic OCR, language models, and AI assistants with an Ethiopian language focus. Built to become part of your product.
Demo-led access. Public API onboarding by request.
The language workspaceIllustration
A thought, in your own words
ሰላም፣ እንዴት ነህ?
Text you can work with
Capture. Understand. Respond.
The product toolkit
From words.
To a useful next step.
Start with one capability. Connect speech, document recognition, generation, and retrieval as your product grows.
Discuss transcription
Text to speech
Give your words a voice.
Explore spoken interfaces, narration, and read-aloud experiences with an Ethiopian language focus.
Research previewLanguage models
Put language to work.
Build chat, summarization, and structured text workflows with configurable model access and generation settings.
Demo by requestAgents and retrieval
Answers with a source.
Connect documents and approved tools to assistants that help people find information and complete everyday tasks.
Demo by requestAmharic OCR
Bring Ethiopic script into text.
Explore extracting Amharic text from images and multi-page PDFs for editing, search, and knowledge workflows.
Demo by requestLanguage coverage, model choices, and deployment scope are agreed for each demo. Voice synthesis is in research preview; commercial availability is not announced.
A place to explore
Meet your next
language workflow.
Explore the interface with prepared examples, then request a live walkthrough.
Interactive previewSample inputPrepared example
A simple greeting
See how a short spoken greeting appears as text.
Illustrative document excerpt
Illustrative input. No audio is uploaded or recorded.ሰላም ለሁሉም።
Interface preview only. Changing this value does not run a model.
Sample transcriptPreview
Your sample appears here.
Choose an example to explore the workflow.This playground uses fixed examples. No live inference, file upload, or microphone access. Request a live demo.
For people who build
Your product.
Your workflow.
A familiar interface for bringing language capabilities into your application.
- Discuss model and language requirements
- Tune generation behavior for your use case
- Agree access, data handling, and usage limits
PythonIntegration preview
import os
from openai import OpenAI
# Details are provided during onboarding.
client = OpenAI(
base_url=os.environ["ETHIONEURAL_BASE_URL"],
api_key=os.environ["ETHIONEURAL_API_KEY"],
)
response = client.chat.completions.create(
model=os.environ["ETHIONEURAL_MODEL"],
messages=[{
"role": "user",
"content": "Summarize these meeting notes."
}],
temperature=0.2,
)
print(response.choices[0].message.content)Illustrative chat integration. Public endpoints, model IDs, and API keys are not issued by this website.
Evidence before a claim
Useful AI should
be measurable.
How would we know if this number were wrong? Our benchmark page starts with the method.
Evaluation roadmap
Results not yet published| Capability | What we will measure | Evidence to publish | Status |
|---|---|---|---|
| Speech recognition | Word and character error rate | Language, dataset, audio conditions | Planned |
| Voice synthesis | Intelligibility and listener preference | Consented samples, native-speaker review | Planned |
| Text generation | Task quality and response time | Model version, prompts, hardware | Planned |
| Amharic OCR | Character and word error rate | Script coverage, scan quality, layout, dataset | Planned |
| Knowledge assistants | Answer grounding and citation accuracy | Source documents, questions, evaluation method | Planned |
What a published result will include
The model revision, language, evaluation date, dataset license and split, hardware, settings, sample size, metric definition, and known limitations. Comparisons will use the same inputs and conditions. We will report uncertainty where relevant and separate upstream research results from EthioNeural measurements.
This is a benchmark placeholder. No accuracy, latency, or superiority claim is made here.
The company
Built here.
Open to what comes next.
EthioNeural develops speech, document recognition, and language AI products with a focus on Ethiopian languages and the people who use them.
We bring open models, careful evaluation, and application engineering together. Our goal is to help teams turn language technology into something people can use every day.
Our current offerings are introduced through demos and scoped onboarding. We build with upstream open models and tools; we do not claim to have trained every underlying model.
Let's build something useful
Bring us your
language challenge.
Tell us what you want to build. We will discuss a demo, model fit, and the path to API access.
founder@ethioneural.comFor other questions:
anything@ethioneural.com