v1.1.8
The kettle had boiled twice before she noticed, and by then the window had fogged over entirely. She wrote the sentence down anyway, because forgetting it would have been worse than getting it wrong.
READY · brief.mp3 · 0:12
brief.mp3 · 0:12 · 48.0 kHz · char timestamps
0.99
REALTIME FACTOR
48.0
KHZ OUTPUT
0
BYTES UPLOADED
There is no server to trust, because there is no server.
VOICE CLONING
One short recording. Any voice.
Give it a few seconds of someone speaking and it reads anything as that voice. You, reading your own drafts. A colleague, reading the release notes. The recording never leaves the Mac, so nobody's voice ends up in a cloud.
reference.wav
0:06 · that is enough
Your voice
v_x4kcm2e
Cloned · English · ready to read
VOICE DESIGN
No recording? Describe it.
Voice design is one sentence. Every amber word is a choice; click them and watch where you land.
A with a accent, reading .
Imogen
v_kigcwfr
Your sentence lands on Imogen, one of the fourteen presets. Every one of them started as a sentence like yours.
THE LIBRARY
Fourteen voices, seven accents
American and British English, plus five accents: Spanish, French, Arabic, Eastern European, East Asian. Two voices each, one woman and one man. Every voice answers to a nine-character id.
Harper
v_wzbib6u
Oliver
v_yvg5vef
Imogen
v_kigcwfr
Walt
v_i3aetnc
Milena
v_p9d8pvk
Kenji
v_dpc32yg
Lucía
v_s2yjwft
Théo
v_bprigm9
Mei
v_4gh5jqq
Andrei
v_q3zm9th
Yasmin
v_w3t6jf9
Diego
v_kxy3qbh
Camille
v_cdhqn7a
Omar
v_hb4myaw
MCP SERVER
Your AI can speak now
The app ships a local MCP server. Claude, Cursor, or anything else that speaks MCP can list your voices, generate speech, and write the MP3 to disk. The same trick the cloud services sell as an API, answering on 127.0.0.1.
OFF BY DEFAULT · LOCALHOST ONLY · ONE TOGGLE IN SETTINGS
you: read RELEASE.md as a calm British narrator, save an mp3
● list_voices
14 voices
● speak · voice_id=v_kigcwfr · captions
Imogen · 3,812 chars
✓ release-notes.mp3 · 3:12 · word timestamps
And all of it stays local. The model, the voices, the audio: on your Mac. The tray up there said 0 bytes uploaded. It meant it.
NO UPLOAD · NO ACCOUNT · NO TELEMETRY
THE APP
Every screen, as it ships
No dashboard, no project tree, no render queue. These are captures of the shipping build on a Mac — the same window you get after the download.
01 / 06
Studio
Paste a script, pick a voice, press Generate. The tray plays the take back and writes it out as MP3.
WHY SWITCH
An ElevenLabs alternative you buy once
Same craft — voice cloning, designed voices, MP3 export, an API your AI agents can call. A different deal: it runs on your Mac instead of in a cloud, for a one-time payment instead of a subscription.
LOCAL
Runs on your Mac
Voices are cloned, designed, and rendered on Apple silicon. Nothing you write or record ever leaves the machine.
ONE-TIME PAYMENT
Paid once, yours for good
One payment covers every voice and all v1 updates. No subscription to cancel, no plan to outgrow.
UNMETERED
No usage meter
Generate as much audio as you like. There is no per-character bill waiting at the end of the month.
QUESTIONS
The things people ask before buying
Is txt2mp3 a subscription?
No. You pay once, and that covers unlimited generation and every update within version 1. The price rises as copies sell, but a license you have already bought is never repriced.
What is limited in the free trial?
Only how much text the app will read at a time. Every voice and every feature works exactly as it does in the paid app, the trial does not expire, and there is nothing to cancel.
How many Macs does one license cover?
Up to five. The key is personal to you, so it moves between your own machines rather than being shared with other people.
What do I need to run it?
A Mac with Apple silicon. On first run the app downloads its speech model, its caption model and a Python runtime — about 5.5 GB, once. After that it works with the network switched off.
Does my text or audio ever leave my Mac?
No. There is no account, no upload and no telemetry. The app touches the network for three things only: that first-run download, activating your license key, and checking for updates. Your text, your recordings and the audio it makes never travel.
Can I use the audio commercially?
Yes. The speech you generate is yours. We claim no ownership of it and never receive a copy, because it is made on your machine.
Whose voice am I allowed to clone?
Your own, or the voice of someone who has given you informed, documented consent. Because the app runs entirely on your Mac we cannot check that for you, which is why the terms put the responsibility on you.
Can Claude or Cursor drive it?
Yes. The app ships a local MCP server that can list your voices, generate speech and write the MP3 to disk. It is off by default and only ever listens on 127.0.0.1.
PRICING
The price climbs as copies sell
One-time payment. Every voice, unlimited generation, all v1 updates. The earlier you buy, the less you pay; that is the whole mechanic.
FIRST 20 COPIES
$29
The launch batch.
Buy for $29COPIES 21 TO 100
$69
Still less than four months of a cloud plan.
COPIES 101 AND UP
$99
The standing price. Still one payment.