Narakeet: reviews and analysis

TLDR: Narakeet is the odd one out in this segment because it does not ask for a monthly fee: you buy minutes, they never expire and there is nothing to cancel afterwards. Its most useful trick is one no speech generator offers either, taking a PowerPoint deck with the narration written into the speaker notes and handing back the finished video. The voices, on the other hand, are not its own and it says so in its own documentation, so quality depends on whichever engine covers your language.

What Narakeet is and how it works

Behind it sits a British company, Video Puppet Limited, incorporated in September 2018 with accounts filed through the year ended March 2026. The product used to carry the same name as the company, had its public beta in late March 2020 and shipped under its current name that October. One person runs it today, though there was a second director between 2022 and 2023. This is not registry trivia: on a tool this small, knowing the company is current and who answers for it is worth more than any about page.

What it does is easiest to grasp from the output end. You give it text and it gives you audio, like everyone else, but it accepts two inputs the others do not. The first is a PowerPoint deck with the narration written into the speaker notes: it reads them, voices them and assembles the video with the slides in sync. The second is a markdown script, with scenes, transitions and segments declared inside the file itself. Anyone who has ever built a training video from a deck understands immediately how much work that removes.

The part that is not its own needs stating plainly, because it is the compulsory question in this niche. Narakeet does not train voices. Its own security documentation names the providers serving the speech: Amazon Polly, Google Cloud, IBM Watson, Yandex, Azure, CereProc, Unreal Speech, Rime and Inclusive Solutions. It is an orchestrator rather than a voice maker, and that is worth knowing before comparing its audio against outfits that do train their own models.

What is its own is everything around that, and it is not a thin layer. Slide rendering, video assembly, the per-second billing model, the automation API with its command line tool and GitHub action. Its founder explained as much to a technical community years ago without dressing it up, describing a headless browser doing the slides and the usual audio and video processing tools underneath. That candour is rare and helps place the product.

What using it is like day to day

The first difference shows up before you use it: you can generate without creating an account. The free one stops at 20 conversions with no commercial use, and leaves out the API, SSML scripting and batch creation, so it is for testing and nothing else. But testing is exactly what to do here, and in your own language, for the reason that follows.

Quality is not uniform, and that conclusion is the hardest thing to find online because nobody has written it out in full. It comes from stitching three sources unaware of each other. A native speaker ran a blind listening test in Turkish with fourteen numbered samples and the brands revealed afterwards, and Narakeet landed among the two worst rated of the group. In Hindi, a video creator calls it acceptable and little more. And the single review on its product listing praises its Chinese performance as beyond what they expected. None of the three settles anything alone, and the Turkish one is written by somebody who warns he only got three responses. Together they sketch something that fits what we already know about the architecture: if the voices come from different suppliers, the result depends on which one covers your language.

What needs care is most of what gets written about this tool elsewhere. Reviews circulate accusing it of over-billing by over-estimating minutes, and its documentation says the opposite, that charging runs per second against the final duration of the file produced rather than against an estimate. Aggregate review counts circulate too that we could not verify at source. None of that is published here.

The other real friction is about automation, and it is specific. The video API only accepts markdown scripts. The PowerPoint conversion, the best thing it does, cannot be automated: you have to go through the website. If your plan was a pipeline turning your team’s decks into video every week untouched, that plan does not work.

Pricing and plans

The cheapest pack is 30 minutes for 6 dollars, paid once, and this is the most honest part of the product: no subscription in the shop window, credits that never expire, and whatever you do not spend today still sitting there a year from now. Above it sit packs of 300, 1000, 2500 and 10000 minutes on the same mechanics.

How that figure gets verified is worth telling, because there is a catch. Its pricing table arrives empty in the document the server sends, with ellipses where each amount should be, and its own script fills them in once the page opens in a browser. That script also does something worth knowing: it does not convert currencies, it swaps the symbol according to the country you arrive from. A visitor from Spain sees the same figure with a euro sign. The number the maker writes out in words on its own page, and therefore the one published here, is 6 dollars.

And a warning for anyone comparing prices across its own pages: its general questions section gives a per-minute range that does not match what the pack table implies. The figure tied to a concrete product is the pack one, and it is the only one quoted here.

One detail almost nobody checks matters here: the paid plan is not just capacity, it is what turns the account commercial. Without paying you cannot use the audio in anything that earns money, nor call the API, nor use SSML, nor generate in batches.

Who it is for (and who it is not for)

It fits anyone narrating in bursts who does not want a live fee: somebody making one video a month, a teacher turning decks into material to watch at home, anyone generating audio for study cards, or anyone needing a single voiceover without opening a commercial relationship. For that profile, paying six dollars once and forgetting about it is precisely the shape they want.

It does not fit anyone whose work lives or dies on voice quality. ElevenLabs is on another level there and the difference is audible. It also does not fit anyone needing fine control over pauses, emphasis and pronunciation across long scripts, which is Murf AI territory, nor anyone producing English corporate training who wants traceable voices, where WellSaid sits. And if your project is an audiobook, the missing performance will tell page after page.

The segment ranking lives at AI voices and the wider map at best AI audio tools.

Alternatives to Narakeet

The honest comparison is not about quality but about model. ElevenLabs, Murf AI and WellSaid all sell a monthly subscription with an allowance, and all three sound better. Narakeet sells loose capacity that never expires and adds an input, the deck, that none of the three accepts. If your problem is that the voice has to move somebody, this is not your tool. If your problem is not wanting one more fee on the card, it is the only one in the segment that answers that.

Frequently Asked Questions

Do I have to subscribe?

Not on its public price list, which is one-off minute packs. The credits you buy never expire and there is no auto-renewal to watch. Its automation documentation does mention subscriptions for commercial accounts, so the exact wording is that there is no compulsory subscription.

Why is the price showing in euros?

Because its page swaps the currency symbol according to the country you arrive from, without converting the amount. The number itself is the same. What the maker writes out in words on its own page is 6 dollars for the 30 minute pack.

Are the voices its own?

No, and it says so itself. Its security documentation names the providers serving the speech, among them Amazon, Google, IBM, Yandex and Azure. What is its own is the layer on top: turning decks and scripts into video, per-second billing and automation.

Does it sound good in my language?

It depends which engine covers that language, and there is no single answer. What is verified is that it placed among the worst in a blind listening test in Turkish and drew praise in Chinese. With free generation available without an account, the only serious answer is testing it against your own text before buying anything.

Can I automate converting my decks?

No. The video API only accepts markdown scripts, and the PowerPoint conversion has to be done through the website. It is the most annoying limitation it carries for a team producing at any volume.