Everyday
The detour, and the reason for it
0:00 / 0:00
الشارع مسكر اليوم على خاطر الأشغال، لازم نعملو دورة طويلة باش نوصلو للخدمة في الوقت.The street is closed today for roadworks, so we have to take a long detour to get to work on time.
Text to speech · Tunisian
A Tunisian voice, not a Tunisian accent bolted onto someone else's language.
Live in beta
The spec
What we can tell you today.
- Parameters
- not yet
- Architecture
- not yet
- Voices
- Built-in, plus your ownGET /api/v1/voices for today's set
- Sample rate
- 24 kHzmono, 16-bit
- Scripts
- Arabic, Tunizi, French code-switching
- Latency
- not yet
- Runs on
- not yet
- Licence
- not yet
- Status
- Live in beta
Dido answers requests today, which is what “live” means here. It went out on the strength of what you can hear above rather than on a finished spec sheet, so the rows without values are genuinely unsettled rather than withheld. They get filled in when they are true, not to make the table look complete.
not yet
Naturalness · human rated
not yet
Intelligibility
not yet
Round-trip CER · spoken, then transcribed back
not yet
Real-time factor
Round-trip is the one that matters: Dido speaks a sentence, our speech model transcribes it, and the two are compared. Nothing here is filled in until it has been run properly, and the clips above are not evidence of a number.
Hear it
Say something in Tunisian.
Building with it? Run the same request in the playground and take the code with you.
Female6 clips
A question
A question asked out of a window
0:00 / 0:00
شنوا الحكاية هذي إلي قاعدة تصير في الحومة؟ الناس الكل قاعدة تتفرج و ما فهمت شيء.What's this business going on in the neighbourhood? Everyone is watching and I don't understand a thing.
Warmth
The softest register the model has
0:00 / 0:00
أمي ديما تقعد تحكي على طفولتها و تحسها ما زالت صغيرة عندي.My mother always talks about her childhood, and she still feels like a little girl to me.
Code-switched
Administration arrives in French
0:00 / 0:00
الprofesseur قاللي ارجع غدوة مع الformulaire متاع الinscription.The professor told me to come back tomorrow with the registration form.
Code-switched
English, the way it actually gets dropped in
0:00 / 0:00
ال iphone الجديد غالي شوية، أما ال camera متاعو خيالية.The new iPhone is a bit expensive, but its camera is unbelievable.
Short
Three words, the shortest thing on the page
0:00 / 0:00
آه، فهمت قصدك.Ah, I see what you mean.
Male6 clips
A story
The longest one, told the way a story gets told
0:00 / 0:00
الطفلة هذيكي إلي قاعدة في الركن، عندها طريقة غريبة في الضحك، تحسسك إلي هي تضحك عليك.That girl sitting in the corner has a strange way of laughing; it makes you feel she's laughing at you.
A question
An invitation shaped like a question
0:00 / 0:00
شنوا رايك نعملو طلة على السوق الجديد؟What do you say we go take a look at the new market?
Calming down
Talking someone down, which has its own tone
0:00 / 0:00
يا سيدي اهدا، المشكلة صغيرة و تحلها بالكلام، ما لازمش تعصب و ترفع صوتك هكا.Listen, calm down. It's a small problem and talking will solve it. There's no need to lose your temper and raise your voice like that.
Code-switched
Two languages for one machine
0:00 / 0:00
الordinateur متاعي عمل crash و ضاعلي الfichier الكل بلا ما حفظتو.My computer crashed and I lost all the files without having saved them.
Code-switched
Reading vocabulary, which arrives in English
0:00 / 0:00
قريت article جديد في ال news على الذكاء الاصطناعي و كيفاش باش يبدل السوق.I read a new article in the news about artificial intelligence and how it's going to change the market.
Short
The other end of the length range
0:00 / 0:00
باهي، نروحو لغادي.Okay, let's go over there.
Why it matters
The flywheel only turns if something can speak.
Collecting audio is the slow, expensive part. A voice good enough to survive round-trip verification turns validated text into new audio, and only what comes back clean is kept. That is the loop Dido closes.
The name
Carthage was founded by a woman who talked her way into it.
She bought as much land as an oxhide would cover, then cut the hide into a single thread. The naming is not subtle.
API access
The endpoint is open.
POST /api/v1/audio/speech · "model": "dido"
Dido has shipped: no waitlist, no access review. Sign in and create a key, shown once and stored hashed, and the first request is a copy-paste from the docs. Free while the beta lasts, 60 requests a minute.
No key yet? You can still hear it in the playground.
Straight answers
What you're about to ask.
Which accent does it speak?
Tunis, to start. The clips above were rendered with the voices called female and male, six each, and both voices say the same seven kinds of sentence so you can hear them against each other rather than take our word for the difference. How many built-ins there are at any moment is the gateway's answer and it has moved more than once, so this page does not count them for you: GET /api/v1/voices lists the set that exists today, and the box above is built from that same list, naming whichever voice actually spoke. Where no model is attached, that is the take that exists rather than the one you asked for. How many more ship depends on how much verified audio each one has behind it.
Can it clone a voice?
Yes, on terms. It needs an account, because a voice belongs to somebody and the permission to use it has to attach to a person rather than to an anonymous request. It needs an explicit affirmation that the voice is yours or that you have the speaker's permission, checked on the server and not only in the browser, so an API key does not skip it. And you can delete it at any time: the voice stops working the moment you ask, and the reference recording goes with it as soon as the voice service confirms it is gone. If that confirmation does not arrive, you are told so rather than told “deleted”, and it is retried. The version that was never going to ship is the one with none of those three.
Where are the numbers?
Not measured yet, so not printed yet. Twelve clips and a box you can type into are worth more than a naturalness score with no methodology under it, and you can judge those with your own ears.