Date: Sat, 9 Aug 2025 09:00:24 -0400
From: Aaron Chantrill <aaron.chantrill@dottywood.org>
To: debian-accessibility@lists.debian.org
Subject: Re: Any Progress on an AI Related Screen-Reader?
Resent-Date: Sat, 9 Aug 2025 13:00:47 +0000 (UTC)
Resent-From: debian-accessibility@lists.debian.org
On 8/8/25 13:04, Aaron Chantrill wrote:
On 8/8/25 10:27, Samuel Thibault wrote:
The documented way: a module specific to the synth, which knows how to
drive it in a more controlled way than just executing a command.
Samuel
I haven't seen a module specific for piper-tts yet. You are saying there
is one? And it should be installed in the /etc/speech-dispatcher/modules
directory?
I did a search and don't see it. Do I need to enable another repository?
$ sudo apt-file search .conf | grep '/etc/speech-dispatcher'
...
speech-dispatcher: /etc/speech-dispatcher/modules/flite.conf
...
Thank you, -Aaron
I do note that flite is on this list. Flite is a very good and lightweight
tts engine based on Festival. Its delivery is rather flat, but it uses the
same set of arctic voices that I downloaded in my piper examples and I still
consider it a huge improvement on espeak-ng, and since there is already a
module for it, it should be fairly straightforward to set up.
I've also seen a lot of comments about how it is inefficient to run a
separate instance of piper for each request, and how it makes more sense to
run a piper service where the onnx model only gets loaded once. I see that
there is an http server included with the piper source code. I'm going to try
running that and using the generic driver to send requests to the instance,
unless there is a better module available.