Which apps let you practise by speaking out loud rather than typing?

Updated: 2026-09-15 · Türkçe

Writing and speaking are not the same skill

In a written exchange you can compose a sentence and delete it, think for ten seconds, even scroll back and fix it. In a real conversation you have none of the three. For most people who struggle socially the problem is not not knowing what to say — it is not finding it in that little time, and that is precisely what written practice skips.

Three differences decide it in practice:

• Time. Speaking gives you a second or two to reply. Writing has no limit, so typed practice never produces the pressure at all.
• Silence. In text, a pause is invisible. In speech, two seconds of nothing is a concrete object, and most people fill it in a panic by pulling the topic back to themselves.
• Pace. The same sentence sounds anxious said fast and calm said slowly. That information simply doesn't exist on a page.

What voice mode can measure that text can't

One thing becomes measurable once you are speaking that cannot be read off a transcript: pace — your speaking speed and your pauses. Whether you are rushing, and whether you are fleeing silence, only shows up here. In Sohbet this dimension is scored in spoken conversations only; in typed turns it is meaningless, so it is left out of the total entirely.

The other eight dimensions — follow-up, open questions, reciprocity, using their name, opening, closing, presence and structure — are judged the same way in both modes; the full list is here.

What the speech layer is actually doing

Speech recognition for everyday conversational speed on an ordinary phone microphone is good enough now that it stops being the thing you notice. What still strains it: speaking very fast, dropping foreign words into a sentence, and background noise from a crowded room. Practising in a quiet room visibly improves the result.

Two layers are involved and they are worth telling apart, because they behave differently without a subscription. Recognition — turning your speech into text — runs through Soniox with a subscription and falls back to Apple's own speech recognition without one, which happens entirely on the device where your iPhone supports offline recognition. Synthesis — giving the character a voice — is likewise a Soniox voice with a subscription and the iOS system voice without.

Apps with a voice mode

Where your voice goes

This is the question to ask about any voice product, and on most product pages the answer is not easy to find. Does your speech leave the device to be processed; if it does, is it kept there; and is it used to train a model? Those are three separate questions.

For Sohbet the answer is that the app never writes audio to a file — not on your device and not on our servers. In voice mode the microphone audio is streamed over an encrypted connection to the speech provider to be turned into text, and what they do with audio in transit is governed by their terms. Your transcripts and evaluations stay on your iPhone only; we hold no copy, and there is no account to create and no email to hand over. The detail is in the privacy policy.

For any other product, the same three questions are worth looking up on their own privacy page. The sentence "your conversations may be used to improve the service" in particular means the audio is being kept.

This page is written by the team behind Sohbet — Sosyal Antrenman. We talk about our own app, including where it falls short; we have no relationship with the other apps listed and earn nothing from them.