Faster than a Whisper, Apple’s speech API’s bring developer promise

Ultimately the value of applied artificial intelligence will be found in useful, efficient solutions that deliver results fast and hallucination free. Apple developers now have access to one such solution, as early testing by MacStories finds Apple’s speech transcription AI is twice as fast as OpenAI’s popular and widely-used Whisper. It also costs a fraction of the price.

This is really promising. Apple uses this for transcription in its apps, such as Notes and in phone call transcriptions. It matters because Apple has made its own native speech frameworks available to developers within mac OS Tahoe. They will now be able to make use of the speech APIs to power their apps.

Faster than Whisper

The test show that Apple’s models work at a significantly faster speed than Whisper, processing a 7GBm 34-minute video file in just 45-seconds, 55% faster than Whisper’s fastest model.

Part of the reason Apple can deliver the results so much more swiftly is because rather than processing speech in the cloud, it processes them on the device. That means of course that not only is processing faster, bit it is also far more secure as the speech never leaves the device.

The takeaway?

While it’s nice to now Apple’s take on Whisper is faster, more efficient, and far cheaper (free) to deploy, it is also a very good signal that as Apple does introduce new LLMs it will aspire to similar degrees of quality. It implies that over time developers on Apple’s platforms will be using best-in-class LLMs to drive software solutions that compete well against others on the market, boosted also by privacy and price.

And that’s got to be a good thing.

You can follow me on social media! Join me on BlueSky,  LinkedIn, and Mastodon.