Speech-to-text and text-to-speech, for free

What is already on the phone

Speech-to-text: a student speaks, the phone writes. Text-to-speech: the phone reads aloud. Both are in every Android phone and every iPhone, in the keyboard and in accessibility settings, at no cost, and both are language-model technology that has improved sharply in the last three years. For a student who cannot write fluently (because of a physical difficulty, dysgraphia, a fractured arm, or because their thinking runs far ahead of their handwriting) dictation changes what they can produce. For a student who cannot read fluently, read-aloud changes what they can access.

Speech-to-text: where to find it

• Android: the microphone key on the Gboard keyboard, in any text field. Google Docs has Voice Typing under Tools on a laptop.
• iPhone: the microphone on the keyboard.
• Windows: Windows key plus H opens dictation anywhere. Word has Dictate.
• Any browser: Google Docs voice typing works in Chrome on a laptop.
• Offline, and the most accurate free option: OpenAI’s Whisper is an open-source recogniser that runs on a laptop with no connection and handles many languages and accents well. Free apps built on it exist for phones. It is slower than the built-in tools and considerably more accurate for non-standard accents.

Where it works and where it fails

The recognisers were trained mostly on recorded speech that is disproportionately American and British English, then other large languages. The consequence is the check question: a teacher with a “standard” accent gets near-perfect transcription; a student with a strong regional accent, or a Nigerian, Indian or Kenyan English that the recogniser has heard less of, gets an error every few words. The same for languages: Hindi, Spanish and Arabic dictation is good; Yoruba, Amharic and most small languages range from patchy to absent.

So: test with the actual student, on the actual phone, before building anything on it. Five sentences. If the built-in tool struggles, try Whisper-based apps, which have wider accent coverage. If nothing works well enough, dictation to a person (a scribe) is still the answer, and the model’s role is nothing.

And treat the transcription as a draft. The student reads it back and corrects it. That correction step is not a burden; it is a reading and editing exercise that a student who could not write at all was previously excluded from.

Text-to-speech: where to find it

• Android: Select to Speak under Accessibility. Highlight, tap play.
• iPhone: Speak Selection and Speak Screen under Accessibility, then Spoken Content.
• Edge browser: Read Aloud, on any web page or PDF. The voices are good and there are many languages.
• Google Docs, Word, most PDF readers: a read-aloud option under accessibility or view.
• The chatbots: voice mode reads what they wrote, which for generated material is the fastest route from text to audio.

The voices are natural enough now that students do not object to them, which was not true five years ago. Coverage follows the same gradient: excellent in large languages, and a small language may have no voice at all, or one that reads with the wrong stress.

A routine that works in a real classroom

One phone, headphones, a shared document. The text for the lesson (the chunked version from the previous lesson) is in the document. A student who needs it listens while the class reads. A student who needs to dictate speaks their answer into the document and then reads it back with the read-aloud, which catches most transcription errors by ear.

The cost is a pair of headphones. The stigma is lower than most teachers expect, particularly if the read-aloud is offered to anyone who wants it.

Recording the teacher

The simplest text-to-speech is you. A phone’s recorder, three minutes of you reading the key text or explaining the diagram, saved to the shared folder. Students who need it play it back. It takes less time than making a worksheet and it works in any language and any accent, because it is yours.

The honest limit

A student who dictates every answer does not practise writing, and a student who listens to every text does not practise reading. These are supports for access, so that the content is not withheld while the skill is built, and they are the right thing for a student with a lasting difficulty. For a student whose difficulty is a skill still being learned, the support is used alongside the teaching, not instead of it, and the decision about which a student needs is a teacher’s, or a specialist’s, and not the phone’s.