FOR AI TEAMS · ARABIC DATA SERVICES

Start with a clear question. Build a useful Arabic data pilot.

Bring a defined Arabic speech or language challenge. We will assess whether a focused annotation, data-quality, or evaluation project can answer it—with explicit linguistic conventions, traceable decisions, and agreed acceptance criteria.

Proposed project-scoped services1.0 · 2026-11-17

PROPOSED PILOTS

Start with the use case.

Volume, timing and cost follow rights, feasibility and reviewer availability checks.

01 / ASR

ASR data & error analysis

Input
Authorized recordings, model output and an agreed transcription reference.
Output
Orthographic transcripts and an error register with explicit normalization rules.
Quality check
Adjudicate reference disagreements; document evaluation units and denominators.
Scope this pilot ↗
02 / TTS

TTS & prosody

Input
Audio samples and their texts, with language variety and context specified.
Output
Agreed annotation tiers and observations on pronunciation, stress and intonation.
Quality check
Listening and acoustic inspection; interpret measurements with their settings.
See the lab example ↗
03 / EVALUATION

Arabic model evaluation

Input
Model responses and a defined task sample without personal or confidential data.
Output
A rubric, reasoned ratings and a register of unresolved cases.
Quality check
Reviewer calibration, adjudication and explicit sampling limitations.
Discuss the rubric ↗

DOWNLOADABLE ILLUSTRATIVE DELIVERY PACK

A useful deliverable includes its context.

JSON and CSV records, a bilingual data dictionary, structural checks report, and SHA-256 manifest. This is a synthetic delivery-structure example; it contains no lab recording or lab-derived measurements.

Download example pack ZIP ↓ Read annotation conventions ↗

BEFORE PILOT ACCEPTANCE

Project readiness checklist

  • Establish rights, consent and permitted uses.
  • Agree secure transfer, retention and deletion.
  • Confirm reviewer availability and variety-specific expertise.
  • Define outputs, acceptance criteria, review sample and revision rounds.
  • Document time and cost assumptions and exclusions.
  • Review the contract and obtain the owner's final commercial authorization.

An explanatory checklist, not order acceptance or legal certification.

Choose the question to investigate

Three possible starting points, subject to feasibility review.

  • Speech annotation: Scope orthographic transcription, word or phone alignment, or a specific prosodic tier. Agree the conventions and acoustic review method before expanding the task.
  • Dataset quality review: Examine a defined sample for schema errors, annotation inconsistencies, missing metadata, and questions requiring linguistic adjudication. Return an issue register with actionable corrections.
  • Arabic AI evaluation: Design a bounded ASR error-analysis, TTS listening, or Arabic language-model evaluation task. Specify the material, rubric, scoring method, reviewer requirements, and limits before execution.

A first pilot with a decision at the end

The proposed starting point is one agreed Arabic variety, one use case, a bounded sample of authorized material, and a defined annotation or evaluation scope. Together we establish the input format, conventions, required expertise, review sample, acceptance method, and exclusions. Sample size, price, staffing, and delivery date are confirmed only after feasibility and rights review. The pilot ends with a recommendation to proceed, revise the scope, or stop—not an automatic commitment to scale.

What the deliverable can contain

A proposed delivery package includes a versioned task specification; the agreed annotation or evaluation output; a schema and metadata dictionary; a quality report stating what was checked and how; an adjudication and exceptions log; and a delivery manifest with file checksums. TextGrid, JSON, and CSV are available formats to scope, not interchangeable representations of every annotation task. Licensing, retention, and permitted uses belong in the written agreement.

Make acceptance measurable

Agree the evaluation unit, reference or adjudication procedure, review sampling, metric, threshold, and treatment of uncertain cases before work starts. Report observed results with their denominators and limitations. Agreement between annotators is not automatically evidence of correctness, and a reviewed sample does not certify every record. Revision rounds and acceptance deadlines should be stated in the project scope.

Price the work you actually need

We use project-specific quotations. Recording effort, accepted audio duration, annotation labor, dataset records, expert consulting, and quality assurance are separate pricing units. Annotation effort depends on the tiers, recording conditions, conventions, and required review; one audio hour is not one labor hour. A pilot can establish a measured productivity assumption before a larger quotation. No fixed market price or minimum contract value is claimed here.

A procurement conversation with clear boundaries

Before a quotation becomes binding, confirm the contracting identity and permitted activity, data rights, privacy and transfer arrangements, scope, acceptance criteria, payment milestones, intellectual property, confidentiality, and contract risks. The owner authorizes commercial commitments. We can discuss requirements now; recruitment, recording access, dialect coverage, reviewer availability, and larger delivery capacity must be established for the individual project.

Inspect the approach before sharing data

Read the example annotation guide to examine the proposed conventions and review workflow. Explore the public Hijazi sample in Arabic PhonoLab and the separately labelled synthetic delivery-format pack. These materials illustrate a method; they are not a completed client project, human-speech validation result, or evidence of ready-to-license datasets. The legal information pages describe the remaining review requirements.

Send a brief, not a confidential dataset

Tell us the organization, intended AI use case, language or variety, task, approximate volume, output format, target date, and acceptance requirements. Use the inquiry route shown on the contact page, or prepare an email to arabicling.admin@gmail.com. Do not attach personal recordings, confidential contracts, credentials, or sensitive data at this stage. An inquiry is a request for discussion, not an accepted order or guaranteed response time.

بيانات ظ · ðˤData.ai

Start with a clearly scoped pilot.

Define rights, deliverables, acceptance criteria, and timing first. Confirm feasibility before committing to delivery.

Contact us