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

Saraiki speech expertise beyond generic datasets.

Saraiki is spoken by tens of millions across southern Punjab, yet written Saraiki barely exists online. Our speaker network is rooted in Multan, Bahawalpur and Dera Ghazi Khan.

Why Saraiki is hard for AI
  • Little written text online
  • Strong dialect variation
  • Models trained on Urdu replace Saraiki words
  • No single spelling convention in daily use
The research challenge

Why models turn Saraiki into Urdu

Most speech models learn from text and audio dominated by Urdu and Hindi, so distinctive Saraiki words and verb endings get 'corrected' into Urdu.

اساں گھر ونجنا اے

"Assaan ghar wanjna ae" (we have to go home). A typical model output: "Hum ghar jana ae", where both assaan and wanjna are replaced by Urdu words. Illustrative example.

Regional coverage

Dialects by district

DistrictVarietyStatus
MultanMultani SaraikiPilot network forming
BahawalpurRiasti SaraikiPilot network forming
Dera Ghazi KhanDerawi SaraikiPilot network forming
LayyahThali SaraikiPlanned
Rahim Yar KhanSouthern SaraikiPlanned
Evaluation design

What we test in Saraiki

01

Dialect vocabulary

Words and verb forms that differ from Urdu.

02

Conversational speech

Natural, fast, overlapping village and family talk.

03

Noise

Markets, tractors, roadside and phone lines.

04

Urdu switching

Speakers moving between Saraiki and Urdu mid-sentence.

Collection protocol

How we record Saraiki responsibly

  1. Orthography firstWe fix a spelling convention (Shahmukhi or Latin) with you before transcription starts.
  2. Consent in SaraikiEvery speaker hears the purpose in Saraiki and signs before recording.
  3. Native reviewTwo Saraiki-speaking reviewers check every transcript; disagreements are logged.
Start small

Commission a Saraiki evaluation.

See how your model handles Multani, Derawi and Riasti Saraiki before your users do.