Fabio Manganiello on Nostr: I've been using the Google Assistant library to power my RPi voice assistants for ...
I've been using the Google Assistant library to power my RPi voice assistants for quite a few years by now, in spite of Google deprecating it a long time ago.
So far I've managed to still get it to work in spite of the deprecation - it just needs a bit of hammering because of some ancient Python dependencies.
But now that the RPi5 is out, the first full ARM64 device of the series, the old builds for ARM32 are no longer working - and it's unlikely that new ones will come out.
So I'm now looking around for alternatives (preferably open, so I don't have to sleep in bed with another evil company that will kill my voice assistants whenever a greedy and incompetent managers decides to pivot to something else), and I'm not seeing much around.
Mycroft and Snowboy are both dead, unfortunately.
Jasper hasn't seen a commit since 2017.
Rhasspy seems a bit more active, but that also hasn't seen activity in 9 months.
Mozilla DeepSpeech was ok when I tried, but it's definitely too heavy, not suitable for small hotword detection models, and probably not suitable for real-time detection in general.
And using OpenAI's Whisper or Amazon's AVS puts me back in the same situation that I want to avoid by leaving the Google Assistant library.
It seems like Picovoice has become more mature over the years (their "contact us for an ARM build if you're a business" business model definitely didn't work i guess), and it looks like literally the only viable option out there.
Am I missing anything? Are there some solid and open, preferably on device, speech-to-text technologies that I'm missing?
So far I've managed to still get it to work in spite of the deprecation - it just needs a bit of hammering because of some ancient Python dependencies.
But now that the RPi5 is out, the first full ARM64 device of the series, the old builds for ARM32 are no longer working - and it's unlikely that new ones will come out.
So I'm now looking around for alternatives (preferably open, so I don't have to sleep in bed with another evil company that will kill my voice assistants whenever a greedy and incompetent managers decides to pivot to something else), and I'm not seeing much around.
Mycroft and Snowboy are both dead, unfortunately.
Jasper hasn't seen a commit since 2017.
Rhasspy seems a bit more active, but that also hasn't seen activity in 9 months.
Mozilla DeepSpeech was ok when I tried, but it's definitely too heavy, not suitable for small hotword detection models, and probably not suitable for real-time detection in general.
And using OpenAI's Whisper or Amazon's AVS puts me back in the same situation that I want to avoid by leaving the Google Assistant library.
It seems like Picovoice has become more mature over the years (their "contact us for an ARM build if you're a business" business model definitely didn't work i guess), and it looks like literally the only viable option out there.
Am I missing anything? Are there some solid and open, preferably on device, speech-to-text technologies that I'm missing?