App with Speech Synthesis Markup Language (SSML) support?

Doitsu · 09-13-2015, 09:56 AM

I occasionally use the iOS TTS voices to listen to English ebooks, and some pronunciations are driving me mad. For example, one of the English voices always reads Mr. as ['em 'ɑ(r)] instead of ['mɪstə(r)] and No. as [nʌmbə(r)] instead of ['nəʊ/'noʊ].

SSML, which is included in the ePub3 standard, would be ideal for handling these issues, but there doesn't seem to be a single app that actually supports it.

Does anybody know an iOS/Android app that supports SSML/PLS/CSS3 Speech or at least allows users to add custom dictionaries that override default pronunciations?

jackie_w · 09-13-2015, 10:16 AM

I don't know about the SSML aspect but the Android Acapela TTS Voices allow a custom dictionary. Not free though.

HarryT · 09-14-2015, 02:42 AM

Trouble is, though, that although it's easy for a human being to tell whether "No." is a negative reply or an abbreviation for the word "number", it's not so easy for a machine.

Eg, consider the following:

"Are there 4 books there?"
"No. 5."

And:

"Beethoven's symphony No.5 is my favourite."

This really isn't a situation in which a custom dictionary will help.

doubleshuffle · 09-14-2015, 03:50 AM

Quote:

Originally Posted by HarryT

Trouble is, though, that although it's easy for a human being to tell whether "No." is a negative reply or an abbreviation for the word "number", it's not so easy for a machine.

Eg, consider the following:

"Are there 4 books there?"
"No. 5."

And:

"Beethoven's symphony No.5 is my favourite."

This really isn't a situation in which a custom dictionary will help.

An excellent example why good machine translation is so damn hard to do. The human brain processes such an awful lot of context.

(Sorry for derailing, Doitsu,)

HarryT · 09-14-2015, 04:12 AM

The type of markup that Doitsu talks about in his original post is really the only way to tackle these issues, but:

a) I don't think any reading app currently supports it.
b) It would mean the publisher having to go through a book and manually insert the markup. I don't think any publisher would find it economically viable to do that. Publishers want to sell audiobooks; they don't care about TTS.

I've never found a TTS system that will correctly cope with something like "I live on St Peter St" ("I live on Saint Peter Street"), but any human reader will understand it. Ivona, for example, will always read "St" as "Street" (presumably because it's more common than "Saint").

It's disappointing, though, that any decent TTS system wouldn't read "Mr" as "Mister" (although it would probably not do well with reading discussions on MR

).

Doitsu · 09-14-2015, 04:45 AM

Quote:

Originally Posted by HarryT

Trouble is, though, that although it's easy for a human being to tell whether "No." is a negative reply or an abbreviation for the word "number", it's not so easy for a machine.

With some simple checks it should be possible to disambiguate the negative reply from the abbreviation of the word number, which is almost always followed by a numeral, e.g., No. 10, Downing Street.
However, initial No followed by a period (or other punctuation marks) and no numerals is almost always a negative reply.

If SSML were more widely available, it'd be relatively easy to search for occurrences of No. followed by numerals and define their pronunciation as ['nʌmbə(r)]. (The default pronunciation should be ['nəʊ/'noʊ] anyway, since in most dialogue-heavy fiction books the negative reply easily outnumbers the number sign, if it occurs at all.)

HarryT · 09-14-2015, 04:47 AM

Quote:

Originally Posted by Doitsu

With some simple checks it should be possible to disambiguate the negative reply from the abbreviation of the word number, which is almost always followed by a numeral, e.g., No. 10, Downing Street.

"Does the Prime Minister live at 11, Downing Street?"
"No. 10, Downing Street."

Sorry, I know that's contrived, but it does illustrate the difficulty. And I do agree that "No" rather than "Number" ought to be the default pronunciation; it's by far the more common.

Barty · 09-15-2015, 01:10 AM

What I don't get is why Ivona correctly reads Mr. Smith as mister Smith but pauses after mister as if at end of a sentence?

HarryT · 09-15-2015, 01:33 AM

Because of the full stop. If it's "Mr Smith" rather than "Mr. Smith" there's no pause. Ivona treats any "." as the end of a sentence and so pauses there. It's not very smart.

gregko · 02-10-2016, 02:16 PM

Hi,
on Android or newer Kindle devices compatible with Android, try @Voice Aloud Reader (Google Play https://play.google.com/store/apps/d...perionics.avar, Amazon App Store http://www.amazon.com/Voice-Aloud-Reader-read-aloud/). Has a speech editor that also supports RegEx expressions for more complex speech replacements. It does know that Mr. is an abbreviation and does not end a sentence, and the user may expand the abbreviation dictionary as well. Support for SSML annotations (where the TTS voice in use supports them too) is under development right now.

Disclosure - I'm the author of @Voice, but the questions posted here are the issues that I specifically addressed in my app.

Greg

09-13-2015, 09:56 AM	#1
Doitsu Grand Sorcerer Posts: 5,584 Karma: 22735033 Join Date: Dec 2010 Device: Kindle PW2	App with Speech Synthesis Markup Language (SSML) support? I occasionally use the iOS TTS voices to listen to English ebooks, and some pronunciations are driving me mad. For example, one of the English voices always reads Mr. as ['em 'ɑ(r)] instead of ['mɪstə(r)] and No. as [nʌmbə(r)] instead of ['nəʊ/'noʊ]. SSML, which is included in the ePub3 standard, would be ideal for handling these issues, but there doesn't seem to be a single app that actually supports it. Does anybody know an iOS/Android app that supports SSML/PLS/CSS3 Speech or at least allows users to add custom dictionaries that override default pronunciations?

09-14-2015, 02:42 AM	#3
HarryT eBook Enthusiast Posts: 85,544 Karma: 93383043 Join Date: Nov 2006 Location: UK Device: Kindle Oasis 2, iPad Pro 10.5", iPhone 6	Trouble is, though, that although it's easy for a human being to tell whether "No." is a negative reply or an abbreviation for the word "number", it's not so easy for a machine. Eg, consider the following: "Are there 4 books there?" "No. 5." And: "Beethoven's symphony No.5 is my favourite." This really isn't a situation in which a custom dictionary will help. Last edited by HarryT; 09-14-2015 at 02:44 AM.

09-14-2015, 04:12 AM	#5
HarryT eBook Enthusiast Posts: 85,544 Karma: 93383043 Join Date: Nov 2006 Location: UK Device: Kindle Oasis 2, iPad Pro 10.5", iPhone 6	The type of markup that Doitsu talks about in his original post is really the only way to tackle these issues, but: a) I don't think any reading app currently supports it. b) It would mean the publisher having to go through a book and manually insert the markup. I don't think any publisher would find it economically viable to do that. Publishers want to sell audiobooks; they don't care about TTS. I've never found a TTS system that will correctly cope with something like "I live on St Peter St" ("I live on Saint Peter Street"), but any human reader will understand it. Ivona, for example, will always read "St" as "Street" (presumably because it's more common than "Saint"). It's disappointing, though, that any decent TTS system wouldn't read "Mr" as "Mister" (although it would probably not do well with reading discussions on MR ). Last edited by HarryT; 09-14-2015 at 04:32 AM. Reason: Typo

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Is there any model that includes speech synthesis?	Moska	Which one should I buy?	15	08-17-2015 04:30 AM
High-speed speech synthesis can be used as an intelligence amplification hack	lirazsiri	General Discussions	13	08-22-2014 05:10 AM
902 PB902 - unable to install new text-to-speech language	szakika	PocketBook	3	01-21-2011 05:01 PM
Tomeraider3 Markup Language (TRML)	konrad	Other formats	1	03-10-2010 03:17 AM
Tr3 Markup Language (TRML)	konrad	Workshop	0	02-10-2010 11:57 AM

09-13-2015, 10:16 AM	#2
jackie_w Grand Sorcerer Posts: 6,212 Karma: 16534894 Join Date: Sep 2009 Location: UK Device: Kobo: KA1, ClaraHD, Forma, Libra2, Clara2E. PocketBook: TouchHD3	I don't know about the SSML aspect but the Android Acapela TTS Voices allow a custom dictionary. Not free though.

09-15-2015, 01:10 AM	#8
Barty doofus Posts: 2,521 Karma: 13036221 Join Date: Sep 2010 Device: Kobo Libra 2, Kindle Voyage	What I don't get is why Ivona correctly reads Mr. Smith as mister Smith but pauses after mister as if at end of a sentence?

09-15-2015, 01:33 AM	#9
HarryT eBook Enthusiast Posts: 85,544 Karma: 93383043 Join Date: Nov 2006 Location: UK Device: Kindle Oasis 2, iPad Pro 10.5", iPhone 6	Because of the full stop. If it's "Mr Smith" rather than "Mr. Smith" there's no pause. Ivona treats any "." as the end of a sentence and so pauses there. It's not very smart.

02-10-2016, 02:16 PM	#10
gregko Junior Member Posts: 1 Karma: 10 Join Date: Feb 2016 Device: Kindle, Android	Hi, on Android or newer Kindle devices compatible with Android, try @Voice Aloud Reader (Google Play https://play.google.com/store/apps/d...perionics.avar, Amazon App Store http://www.amazon.com/Voice-Aloud-Reader-read-aloud/). Has a speech editor that also supports RegEx expressions for more complex speech replacements. It does know that Mr. is an abbreviation and does not end a sentence, and the user may expand the abbreviation dictionary as well. Support for SSML annotations (where the TTS voice in use supports them too) is under development right now. Disclosure - I'm the author of @Voice, but the questions posted here are the issues that I specifically addressed in my app. Greg