I’ve remade, literally every single thing! Everything is now shifted from Python to HTML, JS and CSS, with CloudFlare’s Workers AI as both STT and the main AI assistant! This is not the most stable version yet. I haven’t implemented all the commands, but telling the weather, time, and date, as well as clearing the chat works for now.
But, if you find the AI container annoying, click on it, and it will minimize itself!
Super Huge Update!!!!!
I’ve remade, basically every single thing. Everything is now shifted from Python to HTML, JS and CSS, with CloudFlare’s Workers AI as both STT and the main AI assistant! This is not the most stable version yet. I haven’t implemented all the commands, but telling the weather, time, and date, as well as clearing the chat works for now.
Btw, a container would pop up after you ask the AI. Click the container to minimize it.
Log in to leave a comment
In this update, I’ve added support for macOS, and also persistent memory. It is different from history, which you can tell the AI to save data persistently.
Log in to leave a comment
I’ve replaced the plain STT engine with MicroWakeWord as the WakeWord engine. It’s more accurate, and prevents mis-detecting.
To train the model, you will first have to do “python setup.py”, and then it will guide you through training the model. You can either record “Hey Piston” as the wakeword, or choose your own wakeword. Just record your desired wakeword as it prompts you to.
Btw, don’t start saying the wakeword until “🔴 Recording… press ENTER to stop” pops up
And also, I’ve coded the script to abort the recognition when no one is talking. But if you said something, it would minus one in validCount, so the chat can go on.
Log in to leave a comment
In the previous 4.5 hours, I’ve rewritten the these things:
And yeah, I’ve also fixed the alarm.set.py, and tested to be successful. For AlarmGoOff.py, I’ve made it able to say the purpose (“label”) of alarm, and play a ringtone. Ringtone credit: pixabay.com
here’s an example format of an alarm in data/alarm/alarms.json. Feel free to edit anytime!
Log in to leave a comment
In the previous 7.5 hours of coding, I’ve added an alarm daemon, a weather app, and Piston’s main speech-to-text engine “VOSK” has been replaced by OpenAI’s FasterWhisper.
You can simply tell the AI to list, add and clear alarms in a natural way, e.g. “Wake me up at 8 in the morning.”
You can also ask the AI to tell you the weather like windspeed, temperature, and wind direction in your region.
For the STT model, it is now more accurate, and the recognition doesn’t suddenly stop when you’re still talking, by using the RMS technology to detect silence from the mic.
Log in to leave a comment
Cooool!
Maybe add a dashboard showing those info?
I will try later, but first I need to finish the backend. I literally just messed up on the code execution part lol (subprocess.run())
