Does it really get Polish numbers and units right?

In our test it did: the script said "24 h" and the recording says "24 godziny". We checked that with speech recognition rather than by ear. For comparison, the same shorthand on the MiniMax engine came back as "330 milinów", so this is not something every voice engine does.

Is this the same voice as the one on a product video?

Yes — this is the engine the app uses to put a voiceover under a product clip. A video with a voiceover is billed as one operation on the clip; this page is about the read itself as a file.

How long a script will it take?

This tier publishes no character limit, which is why there is no such row in the table beside it — we do not invent a number the vendor does not state. In practice, split long descriptions into paragraphs and use the before and after fields.

Voice model · ElevenLabs

ElevenLabs Multilingual v2 when the script has clock times, units and delivery windows in it

Nine adverts out of ten for a shop end on a number: what time to order by, how many days to return, how long delivery takes. This engine has a field for text normalisation and it actually performs it — "24 h" in the script came back as "24 godziny", in the right form. That sounds like a detail right up until you have forty product descriptions to expand by hand.

Claim 40 credits See pricing

Rendered by this model

Turn the sound on and listen

–:–– / –:––

The script the engine was given

Zamówienia złożone do 14:00 wysyłamy tego samego dnia. Paczka dochodzi w 24 h, a na zwrot masz 30 dni — bez pytań i bez opłat.

Read in Polish by Sarah with text normalisation on. The script says "24 h" and "do 14:00" and the recording says the expanded forms — we checked with speech recognition, which heard "24 godziny". Put through the same check, the same shorthand on another engine came back as "330 milinów".This was generated by the model this page is about, exactly the way the app runs it. It is not a photo shoot and not client work.

Specification

Built by: ElevenLabs
Speaks Polish: yes — checked on a Polish sentence
Ready-made voices: 20
Steering the delivery: sliders: stability and expressiveness
File from the engine: mp3
Render time: about 10s
Cost: 1 credit per 1000 characters, rounded up

What it is good for

Abbreviations and clock times expand themselves: A separate field turns normalisation on rather than leaving it to the engine to decide. Our test: "24 h" came out as "24 godziny". You do not rewrite shorthand before recording, which across a product catalogue is the difference between an hour of work and one click.
Fields for the text before and after: The engine takes the previous and next sentence, which it does not read — they exist so it knows it is in the middle of a paragraph. On a long description cut into parts, that is the only way the stitched reads do not sound like a list of separate sentences.
Four sliders, not one: Stability, similarity to the voice, style and a speed between 0.7 and 1.2. Set once, they can be saved as the brand read instead of being found again from scratch on every advert.

When to pick something else

It is not the quickest of the ten and it publishes no character limit, so a very long script has to be cut by hand and held together with the before and after fields. If you are auditioning takes and only the wait matters, ElevenLabs Turbo v2.5 has the same twenty voices and the same sliders and comes back sooner. If you want to steer the mood mid-sentence rather than across the whole read, that is ElevenLabs v3. And if you need a pause of an exact length in an exact place, none of these three can be asked for one — MiniMax Speech 2.8 HD can.

See every model

How to brief this model

The things this engine handles differently from the rest. The habits that hold for every model are collected separately, in the help centre.

Turn text normalisation on explicitly rather than leaving it on automatic. It is what turns "24 h" into "24 godziny" — measured on this engine, not promised.
Split a long description into paragraphs and give each one the sentence before and the sentence after. The engine does not read them, but they are why the parts sound like one recording rather than a stitch.
Set the language code to Polish. The engine errors on a language it does not support, so it is a check rather than a decoration.
Keep speed between 0.7 and 1.2 — outside that range the job is rejected rather than clamped to the limit.
Raise stability when you are recording a series in one voice. A low value gives more expression and less repeatability, and across forty products repeatability is what you want.

How to write a brief that works

Questions about this model

Other models

Try this model on your own product

Sign up without a card and get 40 credits. Switching model is one click, so you can send the same brief through several of them and keep the best frame.

Claim 40 credits Work out your saving