Is there a free speech-to-text API?

Yes, you can start with a free speech-to-text API. Each of the four providers below gives a free allowance, and the allowances are big enough for a test. None of them is free at production volume. Once an allowance runs out, the provider bills per minute or per hour, and the open-source option trades that bill for hardware and maintenance. The speech recognition overview explains the underlying technology, and the guide to speech-to-text APIs compares prices in full.

Four providers give a free allowance, and each allowance has a limit

The allowances differ by a wide margin. Google's 60 minutes is one hour of audio ¹. AssemblyAI's pre-recorded allowance is 185 hours ³, far more than Google's or Azure's. For a test, that is the most room, but a sign-up offer can change, so check the page on the day you sign up.

Google and Azure allowances end, and then they bill

Google's free row covers the first 60 minutes each month, and the V1 table bills every minute after that. Data logging changes the price ⁵, so the rate you pay depends on whether you opt in. The Google pricing question explains the logging choice and the V2 table.

Azure's free tier is the F0 plan. It gives 5 audio hours a month of standard real-time transcription ², and batch and other speech features bill separately. If your work is file transcription rather than live audio, the Azure free tier may not cover it.

Open-source Whisper has no per-minute bill, but it needs hardware

OpenAI's Whisper code and model weights use the MIT License ⁶, so you can run them without a license fee. The cost moves to your own machine. The large model needs about 10 GB of video memory ⁷.

Whisper is not a hosted API. You build the service yourself, so the guide to running Whisper locally and the open-source speech recognition guide cover the setup. The hardware and the time to keep the service running are the real costs, and they grow with the amount of audio you process.

The browser's Web Speech API is free, but it is not a server API

The Web Speech API is built into some browsers, and a web page can call it. It is a browser feature, not a service you call from your own server. On some browsers, including Chrome, speech recognition uses a server-based engine, and the audio is sent to a web service ⁸. The Web Speech API guide covers how it works.

We did not use forum threads as evidence

Forum threads on this topic are a common starting point, and we did not rely on them here. Every figure on this page comes from a provider's pricing page, from Google's or Microsoft's documentation, or from the Whisper repository, each read on 2026-10-17. Vendor prices and allowances change, so confirm them on the day you decide.

What to do

1. For a test, sign up for two of the free allowances above and send both services the same 30 minutes of your own audio. The speech-to-text API guide describes the steps.
2. Measure accuracy with word error rate on the same reference transcript for each service, and count any invented sentences separately.
3. If your audio is sensitive, or you expect more than the allowance, test Whisper on your own hardware next. It is the only option here with no per-minute bill.
4. Once you know the volume you need, price it against the rates in whether Google Speech-to-Text is free and what the OpenAI speech-to-text API costs, and read how to use the Google API if Google is your choice.

For dictation rather than an API, the right tool is an app. Yap, which we make, is an on-device dictation app for Mac and Windows. It does not replace the services above for a program that needs transcripts.