FAQ

Yes. The app does not make any network requests. It works fully in airplane mode after install.

No. See the privacy policy — we collect nothing.

No. SpellSynth has no account system, no email signup, no login. You install it and play.

SpellSynth ships with 500+ pre-recorded words across 20+ themes, in a warm retro voice. For any word not in the built-in library, the app falls back to the iOS speech synthesiser, routes the audio through whichever voice character (1968 Odyssey / 1978 LPC / 1986 DECtalk Style / Default) you've picked, and caches the result to disk so the second tap is instant. On-device only; no audio leaves your phone.

Yes, as of 1.2.6. Type a number and it's spoken as words ("2" → "two"), and contractions like "don't" are kept intact instead of being spelled out.

It's now part of Free Play. Free Play has a SPACE key, so you can type a single word or a whole phrase and hear it read back word-by-word — no separate mode needed.

Not in the current version. Spell, Mystery and Say It draw from the built-in word library. Free Play still speaks any word you type.

Yes, as of 1.5.0. Pick FRANÇAIS under VOICE LANGUAGE in Settings and the voice, the words and the game screens all switch over. More in voice language.

Not at the moment. SpellSynth is iOS / iPadOS only. It does run on Macs with Apple silicon via "Designed for iPad" mode.

A 16-step sequencer, added in 1.3.0. Each step holds a word that speaks in your chosen voice — 40 to 240 BPM, on 1/4, 1/8 or 1/16 notes. Type four words and you have a robot vocal loop. Put a single letter on a step instead and it uses the recorded alphabet voice, so you can spell a word out on the beat and then say it whole. FIT stretches each word to exactly fill its step, so words pitch up as the tempo climbs. See SEQ.

Yes, when the host provides one — SEQ follows the host tempo and transport in the AUv3. A host that reports no tempo or transport at all leaves the sequencer running on its own clock at 120 BPM.

In the app, any word. A SEQ step plays from the 500+ word library when the word is in it, and falls back to on-device speech when it isn't — exactly like Free Play. In the AUv3 a SEQ step is narrower than the text field beside it — a step fires on the beat and can't wait for a render, so it plays either a library word or one that has already been rendered, whether by the app or by typing it into the plug-in once. A word it can't find shows a ⚠ badge on that step rather than failing silently. A step holding a single letter always uses the recorded A–Z bank.

Open GarageBand (or AUM, Loopy Pro, Cubasis, Logic Pro for iPad, etc.) on your iPhone or iPad, create a new instrument track, choose "External instrument" or "AUv3 Instrument" and pick SpellSynthAU (listed under nono.audio) from the list. Play notes on the keyboard — each MIDI note triggers a voice sample. The steps are in Loading it.

Yes. The plugin has a LIBRARY / LETTERS mode switch. In Library mode, a text field takes any word and plays it back as a single retro sample — from the built-in 500+ word library, from any word you've already played in the main app, or (as of 1.4.0) synthesised by the plug-in itself. Only a brand-new word pauses to render, showing SYNTH... while it does, and a word the plug-in renders is then instant in the app too. A VIEW LIST link opens the full library. In Letters mode, the yellow A–Z keypad is front and centre — tap a letter and hear it spoken, like the original toy. SEQ steps draw on the same library.

Note On / Off with three velocity curves, Pitch Trigger and Letter Map (notes 60–85 play A–Z) modes, pitch bend up to ±12 semitones, sustain (CC 64), aftertouch routable to any performance parameter, 14 MIDI Learn targets that save with the preset, a tempo-synced Stutter and panic CCs 120, 121 and 123. The full list is in MIDI.

Yes — ten factory presets: Default, Robot Voice, Tape Warp, Telephone, Tempo Stutter, Classic 1978, Hawking 1986, Odyssey 1968, Robot Chant and Spelling Bee. The last two drive the step sequencer. You can save your own via your host's preset browser.

Yes, that was a bug. The bundled voice samples are 24 kHz but the plugin was reading them at the host's rate (typically 48 kHz), which shifted every sound up a full octave. 1.0.5 compensates for the sample-rate mismatch, so the plugin sounds at its intended pitch on any host.