# Typecast: reviews and analysis

> Narration with emotion set line by line, from a maker that trains its own voices.

- Canonical: https://serchai.com/en/reviews/typecast/
- Site: Serchai (https://serchai.com) — AI tools comparator
- Language: en
- Updated: 2026-08-05

---

## Verdict

Typecast is one of the few in this segment where everything points to in-house speech technology and nothing points to a third-party engine: behind it sits Neosapience, Korean in origin, with published research, open source code and two funding rounds covered by outside press. What sets it apart is per-line emotion control, not raw naturalness, where ElevenLabs still leads. Its entry plan is cheap, but the free one is barely a shop window and forces attribution on anything you download.

**Best for:** Anyone voicing scripts with characters or tonal shifts who wants to mark them by hand without studio rates.

**Rating:** 3.6/5

## Pros

- It sets emotion and intensity line by line, rather than one flat voice for the whole script
- Everything points to in-house speech models: published research, open source code and no third-party provider declared
- The entry plan costs 5 dollars a month, among the cheapest in the segment
- It ships an iPhone app alongside the web studio, and updates it often

## Cons

- The free plan is 3,000 lifetime credits, about five minutes, and requires attribution on downloads
- A commercial licence is not part of the free plan: selling what you produce means paying
- The API is billed separately from the web studio, so they are two different subscriptions
- On raw naturalness it does not reach ElevenLabs, and its own iPhone listing sits at 2.5 stars
- Public reputation is thin for its size: there is little user conversation outside aggregators

## Key facts

- Price: Permanent free tier + paid from 5 USD
- Free trial: No
- Platforms: Web, iOS
- Categories: [Audio](https://serchai.com/en/best-ai/audio/)
- Official website: https://typecast.ai

## What the internet says (agentic sweep)

The compulsory question in this segment is whose voices these are, and with Typecast the answer points in-house even though nobody outside has audited it: its maker, Neosapience, publishes a model with a name and version of its own, keeps 86 open repositories holding the official SDK and implementations of its own papers, and no source anywhere names a third-party speech provider. What it sells is not raw naturalness but control over the performance, marking emotion and intensity line by line. The drawbacks are pricing and reputation: the free plan is three thousand lifetime credits, around five minutes, with mandatory attribution and no commercial licence, the API is bought separately from the web studio, and for a tool claiming millions of users there are barely any public reviews where you can read someone describing how it went.

- Sweep date: 2026-08-05
- Derived score: 3.6/5

### Axes

- Results: 3.6/5
- Control: 4.1/5
- Real price: 3.2/5
- Integration: 4/5
- Support: 3.4/5

### Recurring themes in favor

- It marks emotion and intensity sentence by sentence, which is the real difference against generators that apply one tone to the whole script (strong theme)
- Everything points to in-house speech technology and nothing points to a third-party engine: a model with a name and version, papers implemented in open repositories and an official SDK in more than ten languages (strong theme)
- The entry paid plan is 5 dollars a month, among the lowest in the segment, and the figure is also declared in its own page's JSON-LD (present theme)
- The product is alive and moving: repositories with commits from this very week, an iPhone app updated days ago and a funding round closed in December 2025 (present theme)

### Recurring themes against

- The free plan is 3,000 lifetime credits, about five minutes of downloads, and requires attribution on everything you take, so it is for trying rather than working (strong theme)
- The commercial licence and high quality downloads start at the paid plan: without paying you cannot use the audio in anything that earns money (strong theme)
- Public reputation does not match the claimed size: the review aggregators with volume return errors when read and the only openable thing is two ratings on the app store (strong theme)
- The API is subscribed separately from the web studio and its rates cannot be read without running the page's JavaScript, so anyone automating pays twice and with less transparency (present theme)
- The maker itself warns that app store prices may not match those on its site, so the figure depends on where you go in to pay (present theme)

### Sweep sources

- [Official pricing] https://typecast.ai/pricing — Página oficial, leída tres veces de forma independiente el 5 de agosto de 2026. El precio se sirve ya renderizado en el HTML: el plan Basic aparece como «$5» junto a «/month», con el conmutador en «Monthly» por defecto y la línea «$54 billed yearly» oculta hasta activar el anual. El JSON-LD de la misma página lo confirma por su cuenta con «"name":"Basic","price":"5","priceCurrency":"USD"». La escala es Free 0, Basic 5, Plus 19, Pro 29 y Business 69 dólares al mes. Aquí están los límites del gratuito, «Lifetime download credits 3,000 credits (~5 mins)» y «Attribution is required for all content downloaded on the Free plan», y el aviso de que «Prices on the App Store and Google Play Store may differ». El pie escribe «Typecast US Inc. 400 Concar Dr, San Mateo, CA 94402, USA»
- [App stores] https://apps.apple.com/us/app/typecast-ai-voice-maker/id6621190089 — Ficha de iPhone publicada por Neosapience, Inc., actualizada cuatro días antes de esta lectura con la versión 2.5.1. Corrobora desde fuera de la web del fabricante el precio de entrada, porque enumera las compras integradas como Basic 5,00 dólares, Plus 19,00, Pro 29,00 y Business 69,00. Sus notas de versión describen el cambio a un sistema de créditos con un carácter por crédito. La valoración es de 2,5 sobre 5 con solo dos votos, así que no es una muestra y así se cuenta
- [GitHub] https://github.com/orgs/neosapience/repositories — La prueba más fuerte contra la hipótesis del envoltorio hueco, y es comprobable en un minuto. La organización mantiene 86 repositorios, entre ellos el SDK oficial de la API en más de diez lenguajes, una herramienta de línea de órdenes, un nodo de automatización, un servidor MCP e implementaciones de papers propios. Hay commits del 3 y el 4 de agosto de 2026, o sea de esta misma semana. Lo que no hay en ningún sitio es mención a un proveedor de habla ajeno, que es lo que sí aparece en la documentación de las herramientas que revenden motores de terceros
- [Press] https://www.kedglobal.com/startups/newsView/ked202107060013 — Cobertura del Korea Economic Daily de julio de 2021, la más antigua que se ha podido abrir. Sitúa la constitución de Neosapience en 2017, describe la tecnología como capaz de recrear el habla de una persona en otro idioma, y da la primera foto del catálogo, ochenta actores de voz coreanos y veinte en inglés. Cita como cliente a Millie's Library para audiolibros y declara 610.000 miembros y 1.200 millones de wones de facturación en 2020
- [Press] https://techcrunch.com/2022/02/22/neosapience-raises-21-5m-to-use-ai-powered-synthetic-avatars-for-creators/ — TechCrunch, 22 de febrero de 2022, sobre una ronda de serie B de 21,5 millones de dólares que llevó el total levantado a 26,7 millones. Es un hecho distinto del anterior y de otra fecha, no el mismo comunicado repetido. Aporta más de un millón de usuarios en ese momento, 170 actores de voz virtuales en coreano e inglés, clientes con nombre como Hybe Edu, y describe a la empresa como startup coreana fundada por antiguos ingenieros de Qualcomm
- [Press] https://uk.news.yahoo.com/typecasting-ai-voice-neosapience-raises-140000043.html — Sindicación de Deadline del 11 de diciembre de 2025, leída aquí porque el original está tras un muro de pago que devuelve 402. Tercer hecho distinto y el más reciente: ronda de 11,5 millones de dólares, sede declarada en Seúl y San Francisco, y salida a bolsa prevista en el mercado coreano para finales de 2026. Describe la función que define al producto, el análisis del guion línea a línea para asignar emoción a cada frase sin que el usuario tenga que escribir instrucciones


> **TLDR:** Typecast sells something almost nobody else in this niche sells, which is direction. Rather than picking a voice and letting it read the whole script in one register, you mark sentence by sentence which emotion you want and how hard to push it. Behind it sits a company with its own speech research and open code to back that up, not a middleman reselling somebody else's engine. The catch is a free tier that hands you roughly five minutes of downloads for life and makes you credit the tool.

## What Typecast is and how it works

The first thing worth settling in this segment is whose voices you are actually buying, because half the products advertised as voice generators are a text box sitting on top of an Amazon or Google engine. Typecast is not one of those, and it is worth spelling out what that claim rests on. Its maker is Neosapience, incorporated in 2017, Korean in origin and today listing offices in Seoul and San Francisco, with the billing entity registered in California. That history does not come from its own marketing but from outside press covering it at three points years apart. In 2021 a Korean business daily described it working on recreating a person's speech in another language, with a catalogue of eighty Korean voice actors and twenty English ones. In 2022 an American tech outlet covered its Series B, north of twenty million dollars, with customers including the education arm of a music label and a catalogue grown to 170 voices. In December 2025 a third piece reported a fresh round and a planned Korean listing by the end of 2026.

On the question of proprietary voices, precision matters, because this is where the niche exaggerates most and nobody audits it from outside. No independent party has certified that Typecast trains its models from scratch. What you can check in a minute is everything pointing that way: the company keeps 86 open repositories, with the official SDK for its API published across more than ten programming languages plus implementations of its own research, and neither its pages nor the outside coverage name an external speech provider anywhere. Tools that resell somebody else's engine usually say so in their security documentation, because they have to declare where your data travels. There is no such declaration here because there is no such provider.

The product itself is a studio in the browser. You paste the script, choose who says it, and adjust how. That second part is what justifies its existence. Every line of the script is an independent unit you can assign an emotion to and dial the intensity on, so the same character can sound tired in one sentence and startled in the next without chopping the project up and splicing files afterwards. Anyone who has tried to make a conventional generator modulate mid-paragraph knows that is exactly where control ends and resignation begins.

That also defines who it suits. A flat informational script, the kind that describes a product in sixty seconds, needs none of this and any competitor handles it equally well. A script with characters, with a turn, with a dry punchline at the end, does need it, and there the difference is audible. The honest counterweight is that on raw naturalness, measured as whether a distracted listener notices the voice is synthetic, the reference in this segment is still somebody else.

There is a second half of the product to separate out early: the API. Typecast sells it as a service apart from the web studio, with its own subscription, so anyone thinking of dropping narration into an automated production chain is not looking at the same price as somebody working by hand in the browser.

## What it is like day to day

Getting in is easy and the brake comes later. You can generate and listen without limit on the free plan, so the trial phase is genuinely free: you audition voices, mark emotions, compare how one sentence lands at two different intensities. The problem arrives when you want to take the audio away. Downloading costs credits, the free plan carries three thousand of them for life, and that works out at roughly five minutes in total. Not five minutes a month: five minutes and that is the end of it. Anything you do download from that account has to credit the tool, which is not something you will do in a commercial video.

That gap between generating and downloading is what shapes the experience. It is honest, because it lets you test properly before paying, but it also means the free plan is not a plan: it is a long demonstration. Anyone actually producing something, even one video a month, starts by paying.

The credit system has also changed recently. On its iPhone app, updated this same week, the maker describes a move to a simpler rule of one character per credit, after years of metering by download time. That is a change in the user's favour as far as clarity goes, because you can measure a script before generating it, but it also means the consumption comparisons floating around, written under the previous system, no longer apply. If you read a review talking about download minutes per plan, it is measuring something else.

The iPhone app deserves one more note, unflattering as it is. It exists, the maker publishes it, and it updates often, which is a sign of a live product. But its public rating sits at two and a half stars, on two votes. Two votes are not a sample and treating them as a verdict would be dishonest. What they do tell you, and this part is information, is that a tool claiming millions of users has barely moved reviews in the store where it publishes its own app.

That is the most uncomfortable gap here. For its declared size, public conversation from real users is thin. There is press coverage of its funding rounds and there are aggregator listings copying each other, but finding people describing how it actually went for them in open forums is hard. That is not an accusation, it is a fact worth having before you commit, because it means that if something goes sideways you will find little help written by anyone other than the vendor.

## Pricing and plans

The entry paid plan costs 5 dollars a month, and that figure is written out on its pricing page with the toggle set to monthly by default. We checked it twice separately, and it also appears in the page's own structured data, which is the hardest source for a vendor to dress up. Above it sit steps at 19, 29 and 69 dollars a month, and on annual billing the page writes lower equivalent figures itself.

What matters about the entry plan is not the volume of audio but what it unlocks. It brings the commercial licence, and that is the real line: without paying you cannot use the audio in anything that earns money, however much you generated. It also brings high quality downloads. So the question is not how many minutes you need, but whether what you are producing will be published somewhere for profit. If the answer is yes, the free tier was never an option.

Two more warnings about price, both of them the vendor's own. The first is printed on the pricing page: what the app stores charge may not match what the website charges, depending on each store's policies. Check where you are subscribing before treating a figure as settled. The second is the API, contracted separately from the studio, with its own table, and that table cannot be read without running the page's script. Anyone automating ends up paying twice and verifying the price with less transparency than in the rest of the product.

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

It fits people producing performed scripts who want to decide the performance by hand: video channels with character narration, game or animation dialogue, fiction audiobooks, scripted podcasts where the tone shifts inside a single block. For that work, marking emotion per line saves the editing you would otherwise do stitching takes together.

It does not fit anyone chasing the most believable voice above all else, where [ElevenLabs](https://serchai.com/en/reviews/elevenlabs/) remains the reference. Nor does it fit corporate training at volume with team management and brand voices, which is [Murf AI](https://serchai.com/en/reviews/murf-ai/) and [WellSaid](https://serchai.com/en/reviews/wellsaid/) territory. And it certainly does not fit anyone narrating twice a year who wants no live subscription at all: that is [Narakeet](https://serchai.com/en/reviews/narakeet/), which sells loose minutes that never expire. The full segment ranking is in [AI voices](https://serchai.com/en/best-ai/ai-voices/), and the wider map in [the best AI audio tools](https://serchai.com/en/best-ai/audio/).

## Alternatives to Typecast

The useful comparison is not about price, because the entry step is similar almost everywhere, but about what each one lets you control. ElevenLabs delivers the most convincing voice and fine control over pronunciation, but the performance is largely decided by the model. Murf AI is built for team production, with a workflow made for reviewing and re-recording. WellSaid sells traceability and voices licensed for corporate use. Typecast stands apart by letting you direct line by line, which is what you need when the script has characters rather than only information. If your problem is making the voice sound human, look at ElevenLabs. If your problem is making the voice act, this is the one that answers.

## Frequently Asked Questions

### Are the voices its own, or licensed from someone else?

Everything indicates they are its own, and it is worth saying what that rests on. Nobody outside has audited the training, but the company has been in speech synthesis since 2017, publishes its own research with open code, maintains its API toolkit across more than ten programming languages, and nowhere declares an external speech provider, which is something tools reselling other engines do have to do.

### Is the free plan good enough to work with?

No. Generating and listening are unlimited, but downloading burns credits and the free plan carries three thousand for life, roughly five minutes in total. It also requires attribution on anything you publish and includes no commercial licence. It is there to help you decide before paying, which is worth something, but not to produce with.

### How much is the cheapest paid plan?

Its pricing page writes 5 dollars a month for the entry plan, with the toggle on monthly billing. Two caveats: the site itself warns that app store prices may differ, and the API is subscribed separately from the web studio.

### Is it better than ElevenLabs?

On naturalness, no. If the test is whether a listener notices the voice is synthetic, ElevenLabs is still ahead. Typecast wins on something else, directing the performance sentence by sentence, and that advantage only shows if your script needs it. With flat informational copy you would be paying for control you never use.

### Can I automate narration through its API?

Yes, but it is a separate subscription. The maker sells the web studio and the API apart from each other, and the API pricing table cannot be read without running the page's script, so check it yourself before budgeting a project that depends on it.

## Alternatives

- [ElevenLabs](https://serchai.com/en/reviews/elevenlabs/) — Near-human synthetic voices for narration and dubbing. (4.4/5)
- [Murf AI](https://serchai.com/en/reviews/murf-ai/) — Synthetic voiceover studio with video syncing built in. (3.7/5)