Wispr Flow is a masterclass in STT. Apple's solution feels like it's from the last century in comparison. Same applies with Apple's TTS when you have ElevenLabs and OpenAI running laps around it. All I need is for my iPhone to do those things natively at the same quality level (because in Apple's walled garden that's the only way to get them usable everywhere).

▲

jjice 7 hours ago | parent | next [-]

But Apple's uses so few system resources and runs fully on device on newer iPhone models (16+ I believe). It's so efficient. I really enjoy using Handy with Parakeet as the model, but the system resource usage is a monster compared to Apple's (although very good).

Looks like Wispr Flow uses a cloud model [0]:

> Cloud based speech processing infrastructure for 1B users

It gets to be a messy comparison because my iPhone can do STT with no latency pretty well fully on device, but Wispr Flow requires a cloud model, but to be fair, older Apple devices do as well. It's not an apples and oranges comparison, but I think those technical details make this a non direct comparison in a few ways.

For on-device with low system resource usage, Apple's is pretty damn good.

[0] https://wisprflow.ai/post/technical-challenges

▲

RobMurray 4 hours ago | parent | next [-]

Apple's stt has been on-device for a long time now, long before iPhone 16. I haven't noticed any improvements since my first ever iphone 5S. I'm pretty sure wispr flow can use on-device models. I use Voiceink[0] which can use parakeet models on-device and can optionally use cloud models.It's like night and day comparing Apple's to Voiceink. The only advantage I find to Apple's stt is less friction. 3rd party apps just can't integrate as smoothly with the system. There's a gesture to activate Appledictation when Voiceover is on.

	▲	georgel 3 hours ago \| parent [-]
		It's been around and available as an API to devs since at least 2021 in iOS. The problem was even on the best iPhone at that time, I could never get it past ~0.8x speed and after 15-20 minutes the device would heat up so much the display dimmed. For context, I was working on a podcast app with on-device transcription, had to park that idea for years before it got to today's performance.

▲

arijun 6 hours ago | parent | prev | next [-]

Apple runs on-device on older models, too, just wimpier models.

▲

Invictus0 5 hours ago | parent | prev [-]

human resources (my voice and time) are far more valuable than the system resources. going to the cloud is absolutely worth it to prevent a typo

	▲	rhdunn 5 hours ago \| parent [-]
		That doesn't work if you have limited or no connectivity (e.g. on a mountain range). There are also privacy concerns, e.g. a doctor using it to transcribe medical information.

▲

adamcharnock 7 hours ago | parent | prev | next [-]

FWIW - I also really like Wispr Flow, but I moved to running the 'Whisper Large' model locally using Handy (https://github.com/cjpais/Handy), which has been essentially as good, while also having lower latency.

	▲	dceddia 5 hours ago \| parent [-]
		Handy is great. It exposes a bunch of open models beyond Whipser too, and though I haven’t tried too many of them, I’ll throw in a rec for the Parakeet model which feels pretty much on par with Whisper for accuracy and is way way faster.

▲

primaprashant 3 hours ago | parent | prev [-]

I’d say STT is pretty much a solved problem. Everyday there is a new product and can be one-shotted by any current top of the line LLMs. Take a look at this [1]. Apple is just stuck in the past.

https://github.com/primaprashant/awesome-voice-typing