That seems like overkill for what seems to be a really simple task, i.e., driving aplay via the CLI on a rPi. Plus, I would rather not add another piece of hardware to the mix if I can instead do this on the rPi I already have.
Linux server stuff:
As root clone this repo to a working directory (I used the PI's home directory).
- Change your userid to "root" (one way to do this is SU -).
- Use the "git clone" command to copy a github repsitory down to your Linux device. For details see:
git clone | Atlassian Git Tutorial
from a terminal, cd to the directory and run "npm install". This should install all the necessary components.
- This is a command for node.js to install everything.
see if it runs via "node app.js", leave it running.
1 Like
Here's where I'm at on this topic...
With some help from @jtmpush18 (my Linux knowledge is a bit lacking), I was able to get his app installed and working on my rPi, i.e., I can send a text message from his HE device driver to the rPi and hear the speach in headphones plugged into the rPi. Response time is very fast, probably sub-second. The voice is very robotic, but there are other voices that can be used that are probably better. I'm pretty confident that I can feed the rPi line-out signal to a stereo-to-mono converter to a small amp to multiple speakers around the house. I plan on bench testing that part later this week.
There's also other open source local TTS apps available that supposedly have better voices. At some point, I will try to modify the driver to use those apps on the rPi. Also, I want to try something like "Linux aplay" on the rPi from a CLI by passing it the name of a wav file. As I mentioned a couple of days ago, driving wav files may be a better solution for me if I don't have too many to record.
1 Like
If your Pi has a Bluetooth radio, you can potentially use a Bluetooth solution for this. I have used the eSpeak method to send TTS to BT speakers.
3 Likes
I have done a TTS from Hubitat -> Amazon Polly->Mqtt->Linux and/or OSX One thing you will have to deal with is the flakiness of all those processes. For example the power goes out and the raspberry won't reboot (they don't). Or the process on the PI doesn't start or the the Bluetooth speaker is offline when the Pi want's to connect. Even after you learn all about PulseAudio and the newer Pipewire you might still fall into ALSA based commands for your scripts. While my system can work for me, I don't use it because something is always broken when you need it.
Echo Speaks works better, fails less often and the echo dots are very inexpensive.
2 Likes
I don't think that solution meets the "100% local" requirement.
It does.
2 Likes
I do run a Deepspeech derived TTS on a Nvidia 3060 GPU because pi4 wasn't faster enough to be 100% local. I've tried projects that claim the Pi4 can do real time TTS. They didn't for me.
I tested Apple Air Play with a HomePod mini. It worked, but there are issues that I don't want to live with.
It had a several second delay (even with cached messages being played multiple times). I need four or five speakers, so price becomes an issue ($100 each vs $10 each for a dumb wired 3" speaker). I wasn't able to figure out a way to make them all play a message simultaneously. They need power, so speaker placement is somewhat restricted. I want speakers that are inconspicuous, i.e., in-wall or in-ceiling.
I do run a Deepspeech derived TTS on a Nvidia 3060 GPU because pi4 wasn't faster enough to be 100% local. I've tried projects that claim the Pi4 can do real time TTS. They didn't for me.
I only did a few simple tests driving eSpeak from HE rules. They were less than 10 words each, which is similar to what my notifications will be. Response time was subsecond on my rPi 4.
If you want in wall speakers (and price is no problem), then you can go with Sonos amps. Hubitat sees them perfectly, and response time is subsecond.
I.e., you are a bachelor. 
2 Likes
I am not a bachelor.
I just finished an installation with 6 Sonos amps and the speakers throughout the house.
p.s. I just checked eBay and used Sonos amps (first version) can be had for very reasonable prices.
I assume that the Hubitat / Sonos integration will work with the "connect" version of the amp.
2 Likes
There are of course many subjective interpretations of "reasonable" and the ROI of multi-speaker install. 
1 Like
The costs (price and time) go up dramatically for 100% local solutions if you decide espeak isn't good enough.
Sure. A Squeezebox server (raspberrypi, old laptop, etc) and as may network speakers as you want, where a "network speaker" == "cheap computing device, such as an old Android phone, a raspberrypi, etc) + speakers (anything from the 3" speakers that Dell used to include with desktop PCs to nicer powered speakers, easily found on craigslist, ebay, goodwill, etc).
as mentioned if you want alerts (ie water leak on xxxx) on your speakers in multiple locations expecially ones that dont have a hub.. than a cloud based solution like alexa google etc really is you only option...
Some Sonos stuff is on sale... I can get the amp and four in-ceiling speakers for about $2K. That's quite a bit for maybe five less-than-ten-word announcements on a normal day.
Your TTS driver, a rPi 4 (that I already have, so zero additional cost), an amp, four speakers, and associated cables cost me less than $100.
To be clear about why I'm pushing back on solutions that have significantly better audio quality and more future potential... the "bar" I'm trying to reach is what the Elk M1G security panel can do for voice announcements, which except for reliability, is a very low bar. In addition to the TTS/rPi4/cheap solution, I've also figured out how I can use the original Elk M1G setup for announcements initiated from the M1G due to its sensors (which from a reliability standpoint is the best thing to do IMHO), and also initiate announcements from HE due to its sensors (also important, but not life critical). And best of all, the cost is $0 because I already have all the components.
1 Like
Be careful of running all 4 speakers from one amp output. Your amp may die from the low impedance load. I wouldn't try 2 speakers from one output but I'm cautious.
1 Like
Aug '22, me.
Still in use. No issues.
For speech, you can easily just run all the speakers in series; volume drops, but just compensate. The weirdness of series speakers will never happen in his described use case.
