[Marble-devel] Voice Navigation

Dennis Nienhüser earthwings at gentoo.org
Sat Feb 5 16:59:05 CET 2011


Hi,

I looked into voice navigation (speaking out turn instructions, mainly
for the mobile version) into more detail today. I made a
proof-of-concept version (described below) and I'm looking for feedback
before I give that the final touch.

Generally I feel that text-to-speech systems still sound too bad at this
point to be usable for day-to-day operation. Instead I'd like to have
pre-recorded voice samples. That approach works quite well given that
the number of words to speak is rather limited. Ideally we'd ship at
least one complete (english) voice from a professional speaker. If we
don't find one, we could still fall back to replacing voice output with
sounds (taking e.g. some existing KDE ones).

My plan is to re-use TomToms approach -- 59 short voice files -- and
generate the sentences to speak from them. One sentence is either a full
voice file or a concatenation of them when distances are involved (e.g.
TurnRight.ogg + After.ogg + 200.ogg + Meters.ogg). Fortunately there are
a bunch of these voice files out [1] [2] which can be downloaded for
free by users (we cannot ship them though). That way we could ship one
default voice sample set (done by us) and allow users to install others
(e.g. localized ones or ones recorded by them) at their liking. The
voice files need some conversion from the .chk format to be playable
with phonon by Marble, but I could setup some website that handles that.

The remaining work is then:
- Polish the code
- Backport to Marble 1.1 (should keep ABI)
- Find a professional speaker that does a Marble voice sample set for us
  - If we don't find one, ask the community
    - If we still don't find one, use KDE sounds as "voices". Doing that
probably makes sense anyway for people not liking their N900 talk to
them, but still want audible announcements.
- Setup some website on my server (~php frontend to unpacking, oggz-rip
and packing) to convert TomTom .chk files to the Marble format to allow
users to install other voices

Feedback welcome.

Regards,
Dennis

[1] http://www.voice-maniacs.com/
[2] http://www.gps-data-team.com/pda-gps-navigation/links/index.php
(scroll down)



More information about the Marble-devel mailing list