Now booking: Pakistan Regional Voice Evaluation pilots for Q1 2027→
Home / Languages / Urdu
Language · Urdu

Urdu speech data built for real-world variation.

Pakistani Urdu as people actually speak it: regional accents, Roman Urdu, and constant switching into English. Collected with consent, verified by native reviewers.

Urdu at a glance
  • Spoken by well over 200 million people
  • 0.03% of Common Crawl web addresses (Jul 2026)
  • Heavy Urdu–English code-switching
  • Written in Nastaliq and Roman Urdu
Applications

Where Urdu speech data is used

ASR

Speech recognition

Accent-robust Urdu recognition for apps and devices.

CC

Contact-centre AI

8 kHz phone-line speech for call transcription and analytics.

AGENT

Voice agents

Intent-rich prompts for banking, telecom and retail agents.

CS

Code-switching

Natural Urdu–English mixing with switch-point labels.

TTS

Text-to-speech

Native listener evaluation of pronunciation and naturalness.

LLM

LLM evaluation

Native raters judge Urdu answers for accuracy and fluency.

Sample specification

What an Urdu utterance looks like in our data

FieldExample (illustrative)
AudioURD-0203.wav · 16 kHz · mono · 3.4 s
Reference (Roman Urdu)Meeting cancel ho gayi hai, kal reschedule karte hain.
Reference (Nastaliq)میٹنگ کینسل ہو گئی ہے، کل ری شیڈول کرتے ہیں۔
Language spanscancel → en · reschedule → en
Speaker metadataKarachi · age band 25–34 · office · laptop mic

Illustrative example of our format. Live consented samples publish at launch.

Acquisition options

Three ways to work with us on Urdu

01

Custom collection

Speech recorded to your script, domain and conditions.

02

Evaluation

Private Urdu test sets to measure your model.

03

Annotation

Transcription and language tagging of your own audio.

Start small

Request an Urdu pilot.

100 consented Urdu and code-switched utterances with a full error report on your model.