Patronus Transcription Quality Guide

High-quality transcription represents what was actually spoken without silently rewriting the speaker’s meaning. This guide is prepared by the TTA Academy Editorial & Training Team, with expertise in transcription, speech-AI projects, data quality and qualification preparation.

The core principle

Listen for words, hesitations, repeated phrases, incomplete starts and meaningful non-speech events. Do not improve grammar or replace a speaker’s wording because another version sounds more natural. Accuracy means faithful representation with the required formatting.

Preserve spoken content

Keep stutters and repetitions when the guideline requires verbatim transcription. Use digits for numbers and capital letters for acronyms when specified. Do not add speaker labels unless the project explicitly requests them. If a word is uncertain, replay the surrounding phrase and use context without inventing content.

Formatting and non-speech audio

Apply punctuation that makes the spoken structure readable without changing meaning. Silence and background noise are not normally transcribed as words. Meaningful events may use an approved tag such as [laughter] when the guideline includes it.

Quality-control pass

Complete three passes: content accuracy, formatting consistency and final audio comparison. Check repeated words, punctuation, numbers, acronyms, approved event tags and accidental omissions. Submit only after the transcript and audio align.