Can AI transcribe Cebuano?

Cebuano has over twenty million speakers, so the question is fair — and the answer from every mainstream AI transcription engine, ours included, is the same: not reliably. Speech models learn from large volumes of transcribed recordings, and that training data exists at scale for Filipino/Tagalog but not for Cebuano, Ilocano, Hiligaynon, Waray or the other regional languages of the Philippines. No setting fixes a language the model never properly learned.

What happens if you upload pure Cebuano anyway? The engine does its best with the phonetics and returns text that looks like language but reads as noise — Tagalog-shaped guesses, misfired English words, fragments. We'd rather describe that outcome plainly than let you spend an upload discovering it. Where the line actually sits: Tagalog-dominant speech with Cebuano vocabulary mixed in fares noticeably better than pure Cebuano, because the model's anchor language is present.

The practical routes today: for Cebuano-dominant recordings, human transcription remains the honest answer. For the common real-world case — Bisaya speakers code-switching through Tagalog and English in business or family recordings — set the language to Filipino (Tagalog) and expect a heavier editing pass on the Cebuano portions. And if the recording matters historically (oral histories especially), keep the original audio safe regardless of what any transcript says; models improve, and the recording is the artifact.