Speechmatics alternative: API or app?

What Speechmatics is: an ASR engine for developers

Speechmatics describes itself as speech APIs powering voice AI. You wire it into your own product through its API. It does real-time transcription with sub-second latency and batch processing, and you can deploy it as a cloud API, on-device, or on-premises. It covers 55+ languages for transcription and 69 language pairs for AI translation, by its own figures.

The buyers are teams building transcription into something bigger: call-center analytics, live captioning, medical and legal transcription pipelines, voice agents. None of that is a single person trying to answer an email by talking.

Pricing tells the same story. Speechmatics is usage-based, billed per audio hour. The free tier gives you 2,400 minutes — 40 hours — of speech-to-text a month, two concurrent real-time sessions, no card to start. Pro starts from $0.24 an hour of audio and caps at 6,000 hours a month. Enterprise is custom, with on-prem deployment and custom models. That is a meter, and a meter is exactly what you want when you are processing thousands of hours through a product. It is exactly what you do not want when you are dictating a grocery list.

The split: an engine to build with vs an app to use

Here is the line, drawn once, clearly.

An engine like Speechmatics is something a developer integrates. You send it audio over an API, you get text back, and you build the buttons, the UI, the storage, and the billing yourself. It is raw material.

A finished app is something you install and run. Whisper by Remskill is the second kind. It is not a speech-to-text API, SDK, or engine. You cannot build it into your own product, call it from code, or pipe audio through it programmatically. There is no endpoint to hit. It is a desktop application driven by a system-wide hotkey.

One name trips everyone up, so let me get ahead of it. "OpenAI Whisper" — the open-source speech model you can self-host and call as an API — shows up in every Speechmatics-alternative list. That is the developer option. It is not the same thing as Whisper by Remskill, the desktop app I make. Same word, different categories. If you want a model to self-host, you want OpenAI's open-source Whisper. If you want a finished tool to dictate with, keep reading.

If you need an API to build on, here is who to look at

If you are here for an engine, I would rather send you to the right one than waste your afternoon. The genuine speech-to-text APIs in this category — the ones that actually replace Speechmatics for a developer — are:

• AssemblyAI — speech-to-text API with batch and real-time, aimed at product teams.
• Deepgram — low-latency streaming API, popular for voice agents.
• Google Cloud Speech-to-Text — the hyperscaler option, broad language coverage.
• AWS Transcribe — the same idea inside the AWS bill.
• OpenAI's open-source Whisper — self-host the model and run it yourself.
• Gladia — a newer transcription API in the same lane.

All of those are APIs and engines you build into your own code. I am not going to invent accuracy percentages or pricing for them (that is how alternative lists end up wrong — confidently quoting a number from a pricing page that changed last quarter). The point is the category: if you need a meter and an endpoint, one of these is your answer, and Whisper by Remskill is not.

What Whisper does instead: hotkey, speak, paste

Now the other group — the people who do not write code and just want to talk instead of type.

Whisper by Remskill is dictation-first. You press a system-wide hotkey, you speak, and the transcription lands at your cursor in whatever app you are already in. No upload step, no project library, no API to learn. The default hotkey is Ctrl+Space on Windows and Command+Option — a hold-to-talk chord — on macOS. You can change it.

Because it types at the cursor, it works everywhere — your email client, a document, a chat box, a code comment — without anyone building an integration for each app. That is the whole trick, and it is the opposite of an engine. An engine waits for your code to call it. This waits for you to press a key. The first time I demoed it to my wife, I dictated a grocery list straight into a text to her. She replied "great, but you forgot the milk." The app worked. My memory did not.

The multilingual models cover 90+ languages for live speech, and the non-English Whisper models can translate spoken input to English as you go. That is spoken-word-to-English, not the 69-pair text translation service Speechmatics sells — different job, smaller scope, honest about it.

Local and offline: no audio hours, no usage bill

In local mode, Whisper transcribes entirely on your machine. The audio never leaves the device, there is no network call for transcription, and there is no per-audio-hour meter. The whole local pipeline — models, on-device AI cleanup, history, custom words, the hotkey — is free for any signed-in user, with no card at signup.

I want to be fair here, because the honesty is the point. Speechmatics also has a free tier — a generous 40 hours a month — and it also offers on-prem and on-device deployment for developers. So "free" and "offline" are not magic words that only Whisper owns. The real difference is shape. Speechmatics gives a developer an engine they meter and integrate. Whisper gives an individual a finished app with zero integration work and no per-hour bill.

This is the one strong opinion I will spend in this article: per-audio-hour metering is the wrong shape for a person who just wants to dictate. At $0.24 an hour after the free 40, a meter makes total sense when you are running a product through it and need the usage data. It makes no sense when the "product" is you, at a desk, answering email. You should not have to think about a clock running while you talk. A flat app price, with no metering at all, fits that life better. If keeping your dictation off the cloud matters to you, that is the same instinct behind private, on-device speech-to-text.

When Speechmatics is the right tool

I would not switch away from Speechmatics if I were building a product on it. If you need to drop transcription into your own application at scale — a call-center analytics dashboard, live captioning, a medical or legal transcription pipeline, a voice agent — Speechmatics or one of the real API alternatives is correct, and Whisper is not. The same goes if you need strict on-prem data sovereignty for many concurrent sessions, or its 69 translation pairs. Whisper has no answer for any of that. It is a single-user desktop dictation app, full stop. Picking the wrong category here costs you a rebuild, not a refund.

What it costs to just dictate

Whisper's local dictation tier is free for anyone with an account, no payment method at signup. There is no usage clock — you are not billed by the audio hour the way Speechmatics meters Pro from $0.24 an hour. The optional Cloud surface, which uses your own OpenAI key for cloud transcription and web search, sits behind a flat app price rather than a per-minute meter. The current numbers live on the pricing page; the only thing worth remembering is the shape — a flat price for an app, not a meter for an engine.