Your voice

Antiphon

Hear it. Echo it. Speak it.

A podcast, a song or an mp3 you already have, cut into pieces short enough to say back.

In development — the advanced shadowing, with real voices.

The app and the service that cuts a recording are built and run here, at version 0.7.5; the service is being prepared and is not public yet, so nothing is released. Shadowing, dictation, the beat and imports work inside Antiphon and are free; what comes from other apps’ paid parts, such as narrow IPA, is Pro.

Today 0.5.2 · in testing

Hear the beatStressed vowels made longer and louder — 120, 140 or 160 % — in the speaker’s own voice, never re-pitched.
Your subtitles, not WhisperA recording and its .srt, .vtt or .ass: the sentences come from your file, aligned inside each line.
Import from Mimesis and LectioA playlist, a folder or a .zip becomes one lesson; when the words came along, nothing is transcribed.
ShadowingChunks of 2–6 words, ×3 speeds, what you said beside what the speaker said.
DictationThe sentence one word at a time, cut from the real recording.
How hard is this?Words, a level, and the speaker’s pace in words a minute.
Narrow IPA ProFrom a bought Phonema on the same phone; broad IPA is free.

Propart of the paid upgrade, coming

What it will do

Give Antiphon a recording you already have — an mp3, a podcast episode, a song, a video — and it cuts it into pieces you can actually imitate: one sentence, one word, with a slower copy of each. Then the drill is call and response. The piece plays, you say it back into the pause, and Antiphon answers with how close you came.

The comparison is always you against the speaker in that clip — the same sentence, the same voice, each side measured against itself so that it compares pronunciation and not bodies. Nothing is scored against a model speaker or a target. Antiphon holds no opinion about what a vowel ought to be; it tells you whether you sounded like the person you chose.

Antiphon is Greek for “sounding back”: a reader sings, and the people repeat. Shadowing a recording you picked yourself is the whole of its job, and it will stay that small.

What that comes to

A recording becomes something you can repeat. Add a file, a podcast, a recording with its subtitles, or a playlist from Mimesis or Lectio; Antiphon cuts sentence and word clips, and slower copies.
Say it back. The clip plays, you answer into the pause, and you hear the two side by side — with a plain answer about the sounds, the melody and the timing, not a report card.
Your progress travels; your voice never does. Over your own Wi-Fi, to your own Google Drive, or as a file in Documents.

What Hear the beat makes audible

The beat of English is in its stressed syllables. Hear the beat makes their vowels louder and longer so you can hear the pattern you are imitating.

together tə·ˈɡɛð·ɚ

əaˈkʌp‿cupəvofˈtitea

English keeps time on its stressed syllables and lets the rest shrink. Together, as each of four speakers said it (bdl and rms, men; clb and slt, women; the median of nine times each): the stressed ɡɛð is the loudest and 2.0–2.9 times as long as the weak tə before it (bar height is loudness, width is length, in milliseconds). Over 47 such words with a schwa first (312 times said), the stressed syllable was 2.0–2.5 times as long. Syllables as Phonema divides them, times from the corpus’s own phone labels. Read speech, CMU ARCTIC. Below, a phrase in Phonema’s own notation: a and of shrink to ə and əv, and ‿ marks a word running into the next.

What it will not do

Antiphon does not teach. There is no ladder of levels, no course it plans for you, no chart of where a vowel ought to sit, and no score per sound against a reference speaker. That is a different and much larger program — Sentio, the flagship, whose whole claim is that you hear it before you say it. Its first app is in testing, and the vowel map is one piece of it you can already look at.

Under a clip Antiphon writes the basic, broad IPA — syllables and stress, and no more. Phonema writes the narrow notation, the weak forms and the linking. Everything past that line is Phonema's job, and Antiphon calls no other service to borrow it.

Your data

No account, no ads, no tracking. Antiphon will have a server, because cutting a recording into clips and listening to an attempt are both more than a phone can do. The recording you add is sent, cut and not kept: the working copy is destroyed as the lesson comes back to your phone, and swept away within an hour if it never does. The lesson lives on your phone. Your own voice, each time you say something back, is scored and deleted in the same request — never stored, never used for anything else. Add only recordings you are allowed to use.

Get it

Antiphon for Android, once it is built. Cutting and scoring will need a connection to the Antiphon service.
Package com.neurophonetics.antiphon

Google Play — not released yet

Questions, or a recording you would like it to handle:

The rest of the family

Separate apps, each doing one job. None of them needs another — they simply go well together.

Redeem TimeNotice. Limit. Redeem.
MimesisListen. Click. Keep.
PhonemaSee every word, as it’s really said.
LectioRead. Click. Remember.
SentioHear it before you say it.
Semantica Coming soonFind the word that fits.
Pragma Coming soonCatch the hint. Answer in kind.