Transcription
The profession of transcriber has changed exponentially in recent years. Real, word-for-word scribes have become ASR editors in the same way that stenography has made way for digital court reporting. Transcribers rely on ASR output and quickly edit out mistakes and mishears.Verbatim, Baby! uses proprietary software that's designed to run offline and without passing through any APIs, which adds an extra layer of security. Our system centres around Whisper X and Faster Whisper.Your audio is yours alone.
We never use client transcription audio for dataset creation. However, if you're looking to elevate your Whisper raw transcripts, we can craft custom substitution files tailored specifically for your needs.
Raw Text vs Human Editing
The quality of raw machine output is not great for accented speech, especially in Africa. Human editors and proofreaders are necessary for a great quality transcript. Over the course of enough audio, a specialized substitutions file can be compiled to help Whisper write the raw script better, but our suggestion is to always keep humans in the loop for speaker diarization and critical, contextual corrections.Only humans can spot contextual homophone mistakes and variations used by the translation engine for names, numbers, and address spellings. We supply a full glossary along with every transcript linked to sources, and we offer a full array of formats, including ML.
