[arm-allstar] voice announcements

Sat Oct 30 18:42:31 EDT 2021

I got rid of Allison on day 2 of having a working Allstar node.

I wrote a script to do this using a local voice, in my case, one of the 
ones from Mac OS, since it uses bash, and has a pretty reasonable voice  
selection. However, I got slapped pretty good when I suggested ways to 
create and distribute these a couple of years ago. My (very dirty) bash 
script reads line-by-line from a vertical bar delimited text file in a 
while loop, containing the prompt to be spoken on column 1, and the 
output path (relative to current directory) of the file, minus 
extension. It then passes each line separately to the Mac OS Say 
command, including voice name, speed, volume, etc. which writes to .aiff 
format. These are later converted to ulaw with SoX. SoX also trims the 
audio on either side to make it extremely responsive, combined with the 
much faster voice I use. It's about three times faster than Allison. You 
probably wouldn't want something this fast on a repeater, but as a blind 
screen reader user, I'm used to very fast speech.

This script is all pretty unwieldy and unrefined, and I want to do 
something better at some point. I wrote it for my own use, and it 
worked, if only just. The Say command could be substituted for any other 
CLI TTS application, including voicerss, I suppose.

Interesting note: the voicerss American English voices are just renamed 
versions of some of the Microsoft voices that ship with Windows 10. For 
example, John is really David, Mike is Mark, etc.

I'm not sure what the voicerss.org API limitations are for a free 
account, but my hand written text file has 446 entries in it. I included 
some files that I don't think are strictly necessary for app.rpt, 
including all 50 U.S. state names. Asterisk itself has lots of files 
that aren't used by app.rpt, so I didn't generate static files for those.

Now, you might think it's cool to have every prompt in ulaw format. I 
certainly do, because it sounds better. HamVoIP has lots of hard-coded 
.gsm references in scripts, so, on my systems, I have to do a bunch of 
finding and replacing to make it all work. Sure, GSM files are a lot 
smaller, but even people I know who aren't audiophiles can tell the 
difference on an HT speaker.

Here is everything I manually wrote down, and fed to my bash script. The 
third column is just a boundary, because I ran into some weird problem 
with the way I formatted things in the script.
It is assumed that the referenced directories in column 2 preceding the 
file name exist under /var/lib/asterisk/sounds/

On 10/30/2021 12:56 PM, "Dave Petrie via ARM-allstar" wrote:
> Just curious if there is a script or file that is in text format for all
> the voice recordings. My thought is to run a long text file, which I have
> done with no problem on the tts conversion using voicerss.org and change
> the voice and when the conversion is complete and downloaded put the
> separate files back in the proper place in hamvoip. Just for info, my
> duplexer tuning document is 2 pages of text and i converted it in less than
> 15 seconds and then it took 8 minutes and 32 seconds to play, so I don't
> see a problem with making new files, I just don't know how to streamline
> the effort, I understand I could pick and choose the file and replace it,
> that will take forever but can be done.
> I Don't want anyone falling asleep here, that's why I come up with ideas
> now and then.
> Thanks
> Dave
> _______________________________________________
>
> ARM-allstar mailing list
> ARM-allstar at hamvoip.org
> http://lists.hamvoip.org/cgi-bin/mailman/listinfo/arm-allstar
>
> Visit the BBB and RPi2/3/4 web page - http://hamvoip.org