Saudi-dialect voice evaluation data for TTS and voice agents
Native Saudi listeners rate your synthetic voices: how natural they sound, whether the accent is Najdi, Hejazi or simply wrong, which words are mispronounced, and which of two voices they prefer. The scores tell you exactly where a voice stops sounding Saudi.
Specification
- Listeners
- Native Saudi listeners from Najdi, Hejazi, Southern, Eastern and Northern regions
- Tests
- Naturalness (MOS), accent and dialect fit, pronunciation errors, A/B and ranking tests
- Material
- Your synthetic clips, scored alongside native reference recordings
- Annotation
- Scores per clip and listener, marked mispronounced words, free comments in Arabic
- Quality check
- Hidden reference clips and repeat items catch careless listeners
Example of the data
| Voice A | naturalness 4.4 / 5 | accent: Najdi ✓ |
|---|---|---|
| Voice B | naturalness 3.1 / 5 | “الحين” mispronounced |
| A vs B | A preferred | by 8 of 10 |
What it is used for
Choosing a voice
Pick the voice Saudi listeners trust before you launch.
Fixing pronunciation
Get the words and names your voice gets wrong, with how they should sound.
Tracking each release
Score every new model on the same test so progress is measured, not guessed.
Voice-agent QA
Check the whole spoken experience of your agent, not only its text.
Questions
Who are the listeners?
Native Saudi speakers from different regions, so a Najdi listener and a Hejazi listener both judge the voice.
Can you test our own voices?
Yes. Send your synthetic clips; we score them against native reference recordings.
Can I see samples first?
Yes. Request samples with the form on this page and we will send them with the full specification.
Related datasets
Tell us what your model needs.
Request samples and the full specification. We reply within a couple of business days.
Request samples