Hi Karen,
I had described my setup already a couple of weeks ago I think. I used
my BNS emulator, emubns, and connected it to a virtual machine running
SvarDOS. The SvarDOS system has a screen reader that outputs speech in
the BNS format. This BNS speech is received by emubns, who relays it for
actual speech generation to Piper. All this runs virtualized at this
moment, so it is likely not a setup that is desirable for blind people.
However, the exact same result can be obtained by running emubns and
Piper on a Raspberry Pi. In such setup, the DOS system would run on a
hardware PC and output speech to the Raspberry Pi connected via a
standard serial cable. In essence the Raspberry Pi would act as a dumb
speech generator. This, I believe, would be much more user-friendly
since no exotic configuration would be required other than flashing the
Raspberry Pi with a suitable image. I plan to prepare such image, but
have yet to find the time to look into this. Not something very high on
my list of priorities since I am almost sure that I would be the only
user of such system, but still something I'd like to get done eventually.
I believe you use a hardware DECTalk speech generator to make your DOS
talk. To make it simpler to understand, just imagine that instead of
this DECTalk device there is a Raspberry Pi box. The RPi would run
Linux, emubns and piper, but all this would be hidden from "normal"
users, who would consider the Raspberry as a black box that talks when
connected to a DOS PC.
Mateusz
On 19/11/2024 06:18, Karen Lewellen wrote:
Hi Mateusz,
Finally had a moment to check out your piper in DOS sample.
If I am forthright, that was frankly stunning.
granted, I imagine many factors impacted that sound quality, but I have
questions about your process, both how it differs because Piper is in
Linux <what does that mean exactly?> and what you used for the output
for that sample, sound card for example.
There is a discussion of Piper for Linux accessibility, one thing
profoundly missing, speaking personally, is the ease of adjusting basic
things like rate, pitch, inflection, my understanding from command line
Linux users is that those options are not solid. certainly not like the
many DOS screen reader packages.
before I get exacted about a possible safe for my use tts, I am
interested in your process?
Thanks,
Karen
On Wed, 16 Oct 2024, Mateusz Viste via Freedos-user wrote:
On 15/10/2024 17:12, Karen Lewellen via Freedos-user wrote:
Other factor speaking personally with espeak is the largely poor speech
quality.
I experimented a bit with emubns today. Ended up using "piper" instead
of espeak. The speech quality is outstanding now, almost lifelike. I
was able to install SvarDOS using only my ears and fingers and with
the natural voice generated by Piper it was almost a pleasant experience.
Piper is Linux-only, but that's not an issue in the context of running
emubns as a DOS-compatible hardware synthesizer on a Raspberry Pi.
Another problem is that Provox pronounces every period, comma, colon
and parenthesis, which becomes quickly annoying. Perhaps it can be
configured somehow, I have yet to find out. It's all new to me, I have
never used a screen reader before.
I have no idea what the pi would provide speech wise, nor the cost
factor.
With piper integration the speech quality is excellent. The cost
factor is the cost of a Raspberry Pi 3 device and a serial-to-USB
cable. That is, in theory, because I did not test it on real hardware
yet.
Mateusz
_______________________________________________
Freedos-user mailing list
Freedos-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freedos-user
_______________________________________________
Freedos-user mailing list
Freedos-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freedos-user
