handling street names in speech

classic Classic list List threaded Threaded
25 messages Options
12
Reply | Threaded
Open this post in threaded view
|

Re: handling street names in speech

Lester Caine
On 17/07/2019 08:29, Jo wrote:
> Unfortunately TTS is not perfect and it never will be, for one thing
> because it's often difficult to decide with TTS (which language) to use
> for each word in a name string.

For many years I used a TTS engine called Rhetorical on a caller
management system we supplied. It was used to allow calling people by
name rather than ticket number ... we will ignore the privacy debate ...
there was a general consensus that at that time the personal style was
better. The nice thing about Rhetorical was that even without any help
it did a better job pronouncing foreign surnames and some of the staff
did ;) It also had a very good mechanism for adding improvements to
words when there was any particularly obvious mispronunciation. In
addition selection of a different 'output language' worked well,
possibly because the company involved was based in Scotland, and the
clarity of Scottish pronunciation worked well. We had already had to
re-work the older 'English' segmented announcements using a Scottish
voice because of complaints ... but perhaps the main point here is that
this was providing THE SAME English text!

Rhetorical was bought up by an American company and essentially dropped,
but I think CereProc is using the same engine and a large range of
'voices' are available on Android ... if only OSMAnd could access them :(

Text to Speech rendering is exactly the same problem as tile rendering
and adding tagging for particular processes will always be wrong. The
real problem is that there is no mechanism to add corrections in a
secondary system in much the same way as there is no translation system
for the key English elements of the data. Even IPA strings depend on the
context and accent that is being vocalised!

--
Lester Caine - G8HFL
-----------------------------
Contact - https://lsces.uk/wiki/Contact
L.S.Caine Electronic Services - https://lsces.uk
Model Engineers Digital Workshop - https://medw.uk
Rainbow Digital Media - https://rainbowdigitalmedia.uk

_______________________________________________
talk mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/talk
Reply | Threaded
Open this post in threaded view
|

Re: handling street names in speech

ebel
In reply to this post by Andrew Errington
I don't think this counts as “tagging for the renderer”, which is more
about adding false data to “make the map look like what you want” (e.g.
“I want a blue line here, like the `route=ferry` line, so I'll use that”).

I think it could be very helpful for place names which aren't pronounced
the way you pronounce regular English words. e.g. in Ireland: “Dun
Laoghaire” [dun leary], “Tallaght” [tala], “Youghal” [yal], “Portlaoise”
[port leash]. This could be a problem even in England with places like
“Reading”, “Worchester”, “Cirencester”.

On 17/07/2019 00:54, Andrew Errington wrote:

> I think this is a rendering issue (i.e. rendering speech instead of
> graphics) and as such does not belong in OSM.
>
> The work to convert an arbitrary string into speech belongs in the TTS
> engine.
>
> If we start putting IPA strings in OSM then we will start getting
> arguments about the "correct" pronunciation. At the very least it is
> tagging for the renderer, which we should avoid.
>
> IMHO, of course.
>
> Andrew
>
> On Wed, Jul 17, 2019, 09:20 Greg Troxel <[hidden email]
> <mailto:[hidden email]>> wrote:
>
>     John Whelan <[hidden email] <mailto:[hidden email]>>
>     writes:
>
>      > One or two are problematic usually as the street name is an
>      > abbreviation.    For example 1e Avenue in French meaning First
>     Avenue.
>      >
>      > Any suggestions on how these should be handled?  This particular
>      > application is aimed at partially sighted people but I feel we should
>      > be able to come up with a generic solution.
>
>     Two comments:
>
>        osm norms are to expand abbreviations, as I understand it.  So that
>        should be fixed first
>
>        Even after that, we have ref tags, and there is often a road
>     whose ref
>        is something like "CT 2", "US 1", or "I 95".  I don't really think
>        this should be expanded in the database.  Instead, what's needed is a
>        table in the application, perhaps centrally maintained in OSM, of how
>        to pronounce standardize ref abbreviations.  Putting phonetics of
>        "connecticut" on all use of CT or the expanded name is not
>     reasonable.
>
>
>     But I agree this needs help.  I get told to turn on "Court 2" and "Ma
>     2".  Luckily I understand this by now and it actually works ok.  But it
>     does need fixing.
>
>
>     _______________________________________________
>     talk mailing list
>     [hidden email] <mailto:[hidden email]>
>     https://lists.openstreetmap.org/listinfo/talk
>
>
> _______________________________________________
> talk mailing list
> [hidden email]
> https://lists.openstreetmap.org/listinfo/talk
>


_______________________________________________
talk mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/talk
Reply | Threaded
Open this post in threaded view
|

Re: handling street names in speech

Mateusz Konieczny-3
In reply to this post by Jo-2
17 lip 2019, 09:29 od [hidden email]:
Mapping for the renderer means: adding factually wrong data such that it renders the way the mapper wants to see it getting rendered (on the standard rendering).
Or removing correct data to achieve the same.
That's not what adding IPA strings would do.
+1

_______________________________________________
talk mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/talk
Reply | Threaded
Open this post in threaded view
|

Re: handling street names in speech

Greg Troxel-2
In reply to this post by ebel
Rory McCann <[hidden email]> writes:

> I don't think this counts as “tagging for the renderer”, which is more
> about adding false data to “make the map look like what you want”
> (e.g. “I want a blue line here, like the `route=ferry` line, so I'll
> use that”).
>
> I think it could be very helpful for place names which aren't
> pronounced the way you pronounce regular English words. e.g. in
> Ireland: “Dun Laoghaire” [dun leary], “Tallaght” [tala], “Youghal”
> [yal], “Portlaoise” [port leash]. This could be a problem even in
> England with places like “Reading”, “Worchester”, “Cirencester”.

I was about to start off by suggesting that the name->IPA database be
separate as it was not clearly geospatial.  After congratuluating myself
for knowing how to pronounce Portlaoise, I came to "Worchester".  In New
England, we have a "Worcester" and "Marlborough" as well, but they are
pronounced quite differently, "w[oo-shwaish] stah'", and the first R in
marlboro is mostly dropped, boston style.  It definitely sounds funny
when pronounced in straight standard English, and I had already
considered extending osmand to have boston pronunciations.

_______________________________________________
talk mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/talk
Reply | Threaded
Open this post in threaded view
|

Re: handling street names in speech

Colin Smale
In reply to this post by ebel

On 2019-07-17 10:44, Rory McCann wrote:

I don't think this counts as "tagging for the renderer", which is more about adding false data to "make the map look like what you want" (e.g. "I want a blue line here, like the `route=ferry` line, so I'll use that").

I think it could be very helpful for place names which aren't pronounced the way you pronounce regular English words. e.g. in Ireland: "Dun Laoghaire" [dun leary], "Tallaght" [tala], "Youghal" [yal], "Portlaoise" [port leash]. This could be a problem even in England with places like "Reading", "Worchester", "Cirencester".

And homographs like Gillingham (Kent) and Gillingham (Dorset) which are pronounced differently (soft vs hard G)
 
Oh what a wonderful language...
 
 

_______________________________________________
talk mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/talk
12