humanTTS^BETA

Tired of AI voices that sound like robots?

So were we. While the rest of the world races to generate more synthetic speech, we asked a different question.

What if we didn't?

humanTTS is the world's first (and only) fully transparent, deliberately inefficient, non-synthetic text-to-speech API. You send us a string, Myra's phone lights up, Myra reads it into her microphone, and you receive a .wav file. That's the whole product.

It's slow, it's expensive, and it's inconsistent, but it's genuinely human, which is something no amount of transformer layers can give you. We call it artisan speech synthesis, because the word "bespoke" is already taken.

How it works

Four steps. Two humans (you and Myra). Zero GPUs.

1. 1

   You send us text

   POST /v1/speak with your string, a voice ID, and your API key. Your request enters Myra's queue.

2. 2

   Myra does her best

   Myra's phone lights up. She reads your text aloud, attempting the voice you requested. She is not good at accents. This is priced in.

3. 3

   ???? ????????

   Classified. Not even Myra knows what happens in this step, and she is the one doing it.

4. 4

   Audio returned

   High quality audio returned directly to your endpoint!

Voices

Every voice in our catalog is Myra. We believe in consistency.

Voice ID	Description
en-US-Myra	Standard American. Sounds exactly like Myra, because it is Myra.
en-GB-Myra	British. Sounds like Myra pretending to be British. She is from Missouri.
en-AU-Myra	Australian. Sounds like en-GB-Myra but with less confidence.
fr-FR-Myra	French. Mostly "bonjour" and apologies. Do not use for full sentences.
es-MX-Myra	Spanish. Based on two years of high school Spanish and a 41-day Duolingo streak.
de-DE-Myra	German. Currently unavailable. Myra does not know any German.
ja-JP-Myra	Japanese. Four-plus years of high school and college Japanese classes. Honestly just gibberish.
it-IT-Myra	Italian. Mostly gestures, which we unfortunately cannot transmit over HTTP.
en-PIRATE-Myra	Pirate. Not available from any other provider. Arrr.
en-GOTH-Myra	Goth. Same voice, read in a slightly sadder tone.

Pricing

Our pricing is inversely based on how long Myra takes to get to your request. The longer you wait, the less you pay, down to a floor of $0.001 per character. We call it Latency-Adjusted Billing^™.

Response time	Rate	What Myra was probably doing
Under 5 minutes	$0.10 / char	Myra had her phone in her hand.
5 to 30 minutes	$0.08 / char	Myra finished what she was doing first.
30 minutes to 2 hours	$0.05 / char	Myra was in the shower, or thinking.
2 to 8 hours	$0.02 / char	Myra forgot, and then remembered.
Overnight	$0.005 / char	Myra saw it in the morning, with coffee.
More than 24 hours	$0.001 / char	The price floor. Myra did her best. Consider this a tip.

All prices in USD. Volume discounts available for extremely patient customers. Enterprise customers receive Myra's personal cell number and a promise she'll "try harder."

Error codes

Our error responses are fully descriptive and occasionally apologetic. Most of them are Myra's fault, and she wants you to know that.

200	OK. Myra got to it.
202	Accepted. Myra's phone has buzzed.
401	Unauthorized. Myra doesn't know you.
403	Forbidden. Myra has chosen not to say that.
408	Request Timeout. Myra's phone was on silent.
418	I'm a teapot. Myra is making tea.
422	Unprocessable Entity. Myra tried but it didn't sound right.
429	Too Many Requests. Myra is doing her best.
451	Unavailable For Legal Reasons. Nova is in the room.
503	Service Unavailable. Myra is asleep.
507	Insufficient Storage. Myra's voice is tired.
418½	Myra Has Questions. Your string confused her. Reply in the dashboard.

What people are saying

Frequently Asked Questions

Is this a joke?

Everything about this service is real, except possibly the British accent.

What is your uptime SLA?

Myra sleeps approximately seven hours a night, so our SLA is roughly 70%. We do not guarantee availability during showers, meals, or parenting.

Can I get a refund if Myra mispronounces my text?

No, but we will happily re-record it with more feeling.

How is this different from calling a friend who owes me a favor?

We have documentation.

Do you use any AI at all? At any point in the pipeline?

We are radically committed to not using AI for speech synthesis. We did try coding the rest of it by hand as well, but it got super freaking boring, so the API gateway, the billing system, the notification pipeline, and probably most of this FAQ are AI-assisted. The voice is the only artisanal part. The voice is the product.

Do you cache phrases that get submitted repeatedly?

Yes. Once a phrase has been submitted three times, we cache Myra's recordings of it and serve them instantly at a flat rate of $0.04/char. We store up to five takes per phrase and deliver them in random rotation, so your "Welcome back!" doesn't always sound exactly the same. Myra can add fresh takes whenever she's bored. See the cache API docs for how to opt out if you need every response to be fresh.

What about GDPR?

Myra forgets things constantly. We consider this a feature.

Can I hear a sample before signing up?

Yes. Submit any text through the free tier. You will hear Myra do it when Myra hears you do it.

Is there a webhook for delivery?

Yes. It fires when Myra taps "send." Sometimes she forgets to tap send, in which case the webhook fires the next day, alongside an apology.

Sign up for the beta

We're currently onboarding new developers on a strict first-come, when-Myra-feels-like-it basis.

Join the beta

By signing up you agree that Myra is a human being with a life, a daughter, and occasional dental appointments.