Hello Team,
I am trying to integrate Voice TTS +STT to my Raspberry PI. I was researching this topic on YouTube and can’t find a good solution to this problem (based on my skill set). I was leaning towards Mycroft, but they are no longer in business. I don’t want to use Amazon or Google. Since you guys have more experience than me using Raspberry PI and the associated add on, if you can, recommend several alternatives to use that would be great (been at this for weeks).
To recommend something it would be great to know where you’re at in terms of maker skills.
For instance, have you programmed before? Have you done much hobbyist electronics?
Also I’m wondering what audio input/outputs you’re hoping to use? A microphone? Speakers?
I’d need to know what services you are ok with employing.
Concerning not using amazon and google, that’s totally cool.
Are you speaking broadly, i.e. “I don’t want to use a subscription service from any big AI based company.”
Or, are you speaking about those two companies specifically, i.e. “the two companies, Amazon and Google, do not align with my ethics and I don’t want to financially support them”
Pixmusix,
Thank you for your response. My maker skills are in its beginner stage. My background is in Mechanical Engineering, and I don’t have much programming or electrical/ electronic skills. But I did manage uploading the Raspberry OS systems and have installed Magic Mirror on my monitor.
I want to use a microphone and speakers for my project. My project is to have a Magic Mirror with Voice Recognition. My goal is to have Voice Recognition systems with my Magic Mirror at a very basic level. A very short script with very few capabilities.
Example:
Wake Word: Hey Alice:
Hey Alice, what day is it? Hey Alice, what time is it: Hey Alice, what is the weather forecast for today?
Thank you for your help, much appreciated.
Are you willing to pay for subscription services to 3rd parties (excluding google & amazon)?
Hey @TOMMY297623 ,
Have you found DeepSpeech in your search for a suitable voice recognition solution?
It does rely on Google’s TensorFlow on the backend, which may not be okay depending on how much involvement you are happy for Google to have, but otherwise, this seems like a decent, open-source option for voice recognition on a Raspberry Pi.
The documentation for the project is also quite good. You can find that at the following link:
Hope this helps!
yes
Yes it does. Thank you very much.
Look in terms of ease of setup and good modern results I think deep-speech really has it down. If your new to programming, googles documentation is good and beginner friendly.
@Samuel suggested that and you could argue it’s a good blend of Open Source to propriety.
As an alternative check out Kaldri. It’s not promising an easy journey but it’s doable and very debian-esk (which is what raspberry Pi is built on). Good results 
I have a couple of Raspberry Pi-based voice satellites still using Rhasspy software linked to Home Assistant. Mike, the guy behind Rhasspy, worked for Mycroft and now for Nabu Casa (the company behind Home Assistant) doing a wonderful job of developing a fully local voice assistant.
Mike developed Rhasspy as a generalised voice toolkit - but it hasn’t got much attention in the last couple of years. The modules he has developed since (Rhasspy v3 morphed into Wyoming) are designed to be general use … but the focus has been more on integrating into Home Assistant than documenting how to use them stand-alone.
I am interested in Home Automation rather than AI, so these are my limited experiences/opinions and not to be taken as fact.
From a RasPi hardware point of view, we commonly used reSpeaker 2-mic HATs from seeed (and several copies including the Adafruit Voice Bonnet) - but seeed didn’t put much effort into the drivers and software; and then stopped supporting them. I mean, the device proudly boasts 2-mics … and the hardware for both mics does work - but the supplied driver only uses one 
One user kindly has been updating the driver for later kernel releases - but there has been no active development for many years.
Since then seeed released 4-mic and USB reSpeakers … but they never caught on (in the Rhasspy community anyway).
Certainly you can add microphone and speaker to your magic mirror RasPi … but you won’t get good quality without adding Digital Signal Processing (DSP).
Recently I see several ESP32-S3 based voice satellites utilising a XMOS DSP chip; including the ReSpeaker Lite (which claims to work with an ESP32 or Raspberry Pi). I am curious to try it .. real soon now.
As for the back-end … RasPi is a great general-purpose computing platform - but voice recognition (STT and TTS) really benefits from more specialised AI-style hardware. Google and amazon do it “in the cloud” and don’t want you think about the implications thereof. They have spent years and millions of dollars on high powered hardware and AI software to make it look easy.
The fact is that it requires millions of calculations, similar to those graphics cards do to generate the reflections on waves in your latest game. You can get slow low-quality responses on a RasPi - but it is preferable for your Raspberry Pi to pass AI requests to one of the cloud-based services (with the massive hardware and quick response times) - or to set up your own PC with modern graphics card to do the AI.
If you are wanting it mainly to control your home (“open the blind”, “turn off the light”, “Is the garage door closed?”) I would definitely recommend you look at Home Assistant and “year of voice” (now in its second year 
). Recently they have been adding general AI capabilities (including to anonymously ask Alexa or Google) where Home Assistant can’t provide the answer.
Personally I have just purchased a Home Assistant Voice Preview Edition device (ESP32-S3 microcontroller with XMOS Digital Signal Processor, 2-mics, speaker and status LEDs for Aus$120 with delivery) for my Home Assistant … and next job will be to upgrade the software in my RasPi units to bring them into the same Wyoming system.
Thank you everyone. I got some great suggestions. I would investigate all of them and select the best one that fits my current skill level.
And you can get our latest projects and tips straight away by following us on:
