Glossary
Auto-generated captions vs speech-to-text transcripts
Updated · By the WithTranscript team · How we checked
Auto-generated captions vs speech-to-text transcripts: Auto-generated captions are the subtitles YouTube's own speech recognition adds to a video; a speech-to-text transcript is made separately by running the audio through another recognition model.
YouTube creates automatic captions with speech recognition for many videos, in a growing list of languages. YouTube itself says their quality can vary with the audio, accents and background noise, and the creator can edit them. Some videos instead carry captions the creator uploaded or typed, which are usually more accurate.
Most free "YouTube transcript" websites read one of these caption tracks and reformat it. That is fast and cheap, but the result can only be as good as the captions, and a video with captions turned off or not yet processed has nothing to read.
A speech-to-text transcript downloads the audio and runs it through a separate model (Whisper, AssemblyAI, Deepgram and others). It works when there are no captions, can add punctuation, paragraphs and speaker labels, and is often more accurate on hard audio, but it costs money and takes longer, which is why tools that offer it usually charge or cap it.
When you compare tools, the first question is which of the two a tool uses: it decides whether videos without captions work at all, and how much the text needs cleaning before you quote it.
How WithTranscript handles it
WithTranscript builds new transcripts from the video's YouTube captions, then adds paragraphs, chapters and speaker changes on top. Some older transcripts in the library were made with our own speech-to-text. Videos without a usable caption track get no transcript.