Real video, careful listening, producing language yourself, speaking along and recalling words: five ideas from language-learning research, and what DictaBooth does with each.
Standard YouTube auto-captions often fail on fast native speech, regional accents, slurred contractions and background audio. For a learner a wrong transcript is worse than none: it can teach the wrong grammar and hide what the speaker actually said.
DictaBooth's transcript editor reads each transcript against the audio, corrects the errors and checks its own corrections before you practice. Every video also tells you where its words came from (creator captions, automatic captions, or written from the audio), so you know how far to trust them.
For language educators: we check transcripts for teachers and creators free of charge. Ask for one.
The scienceStephen Krashen's input hypothesis says language is acquired from input just one step above your current level: “i + 1”. It is influential and still debated.
In DictaBoothStories come in three levels, so the reading stays within reach, and the material is real unscripted YouTube rather than textbook audio.
The scienceWriting down what you hear forces you to connect spoken sounds to written words. Studies of captioned video find it helps listening and vocabulary (Montero Perez, Van Den Noortgate & Desmet, 2013, a meta-analysis of eighteen studies, doi:10.1016/j.system.2013.07.013).
In DictaBoothYou type what you hear, covered subtitles stop the text answering for you, and every word is marked, so weak forms and fast reductions (“should've”, “gonna”) show up as exactly what you missed.
The scienceMerrill Swain's output hypothesis holds that listening alone does not build grammatical precision: being pushed to produce language makes you notice what you cannot yet say.
In DictaBoothParaphrase plays a part of the video, then asks you to say it your way. Each part of the original turns green, yellow or red for whether your version still says it, and the grammar is checked clause by clause, without stopping your flow.
The scienceShadowing means speaking along with a native recording, which works on rhythm, stress and intonation as well as individual sounds. Small studies report better pronunciation after shadowing practice (Foote & McDonough, 2017, doi:10.1075/jslp.3.1.02foo) and better listening for lower-proficiency learners (Hamada, 2016, doi:10.1177/1362168815597504).
In DictaBoothYour take is scored by Microsoft Azure Speech AI for sounds, speed and pauses, and your intonation is drawn against the speaker's so you can see where the two part company.
The scienceThe testing effect (Roediger & Karpicke, 2006) is the finding that pulling a word out of memory sticks better than looking at it again. Hearing it, spelling it and seeing it in a sentence gives the memory more to hold on to.
In DictaBoothWords you save keep their sentence and the moment they were said, and come back on a schedule as drills: listen and type, fill the gap, say it.
Instead of textbook audio recorded in a booth, DictaBooth uses YouTube, the richest source of living language there is: real conversations, documentaries, accents and the subjects you actually care about.
DictaBooth uses Microsoft Azure's pronunciation assessment instead of a vague percentage. It looks at:
The studies above are about the methods, mostly in small groups of learners, not about DictaBooth. We have not run a study of our own, and a study of shadowing is not a study of a website with a shadowing button in it. What the research supports is that these are reasonable exercises to offer, and that is the claim we are making.
Krashen, S. (1982) Principles and Practice in Second Language Acquisition. Swain, M. (1985) Communicative competence: some roles of comprehensible input and comprehensible output in its development. Roediger, H. & Karpicke, J. (2006) Test-enhanced learning, Psychological Science 17(3).
New to the app? How it works walks through each exercise, and the questions page covers where a transcript comes from and how far to trust it.