Portuguese Transcription Services

Carioca, Paulistano, Nordestino, Mineiro, European Portuguese, Azorean.

Lisbon speech reduces unstressed vowels so heavily that syllables disappear and consonant clusters appear that the written word does not contain. Brazilian speech keeps those vowels. The asymmetry is familiar to anyone who has worked across both: Brazilian listeners struggle with European audio far more than the reverse, and the errors come out as deletions rather than as substitutions.

That is why the markets are staffed separately instead of pooled under a single language name. A Portugal batch goes to annotators in Portugal, a Brazil batch to Brazil, and the review pass stays inside the same pool rather than crossing back over the same gap.

The orthographic agreement pulled the spelling systems close together, which makes the remaining differences easy to miss. Continuous action is written estou a fazer in Portugal and estou fazendo in Brazil, address splits between tu, você and o senhor by market, and pronoun placement moves. These are transcription decisions, not typos, and a reviewer from the wrong market will correct them into a variant nobody in the recording speaks.

Inside Brazil the spread is wide enough to matter on its own. Carioca speech palatalizes coda s where Paulistano does not, Nordestino vowels sit in different places, and conversational pronoun use departs from what the written standard predicts: a gente carries the first person plural while the verb stays singular, and object pronouns land in positions a grammar book rules out. Verbatim delivery keeps those forms rather than repairing them into schoolbook Portuguese.

Portuguese coverage on this page is Brazil and Portugal. Angolan, Mozambican and other African Portuguese speech is real work with its own vocabulary and its own contact features from local languages, and Spirelight is not claiming a staffed pool for it here. If that is your requirement, raise it at scoping and you will get a straight answer instead of a Brazilian pool quietly doing the work and the transcript reading as though it fits.

The published dataset coverage lines up with the staffing: Portuguese (LatAm) and Portuguese (Western) are separate collections for the same reason the annotator pools are separate, and the accuracy target for each is measured on a held-out slice of your own recordings from that market.