Native-speaker transcription, translation, diarization and language tagging for Saraiki, Pakistani Punjabi and Urdu, including the code-switched and noisy speech other vendors avoid.
Every word as spoken, including hesitations and false starts, in agreed script (Latin, Shahmukhi or Nastaliq).
Who spoke when, with word or segment timestamps for training and alignment.
Each span tagged Urdu, English, Saraiki or Punjabi for code-switching research.
Meaning-faithful English translations reviewed by bilingual editors.
Native raters score model answers for correctness, fluency and cultural fit.
We follow your label set and guidelines, or help you write them.
We write dialect-specific guidelines before work starts and log every reviewer disagreement.
| Case | Our approach |
|---|---|
| Overlapping speakers | Separate segments per speaker; overlap flagged |
| Dialect variants | Spelling convention fixed per project; variants mapped |
| Code-switching | English words kept in English script; switch points tagged |
| Unclear audio | Marked [unclear] rather than guessed |
| Names and numbers | Normalisation rules agreed up front |
A second native reviewer checks every transcript, or a sampled share agreed with you.
Disagreements go to a senior linguist; the decision and reason are logged.
Each delivery includes measured error rates from an independent sample.
Illustrative example of our format. Live consented samples publish at launch.
Share 10 minutes of audio and your guidelines. We return a labelled sample so you can judge quality first.