E-book specification supporting multiple voices for TTS

[richough3]

As much as I enjoy reading, I also find I like doing text-to-speech as well, so I was thinking it would be interesting if an ebook spec would also support multiple voices for text-to-speech, so if you had a program read it to you, it could do different voices for the narrator and characters. Of course, I know Audiobook creators wouldn't want something like that, but it still seems like a neat idea.

[JSWolf]

Originally Posted by richough3

As much as I enjoy reading, I also find I like doing text-to-speech as well, so I was thinking it would be interesting if an ebook spec would also support multiple voices for text-to-speech, so if you had a program read it to you, it could do different voices for the narrator and characters. Of course, I know Audiobook creators wouldn't want something like that, but it still seems like a neat idea.

You would first have to create the TTS program to handle certain span styles in the eBook to know to switch to a different voice.

I do like the idea. But I don't know of an existing TTS that would do this.

[Quoth]

It would have to be coded into the ebook. It would be beyond existing parsing for a text to speech to do much more than dialogue and narration. Even that would fail with Angela Thirkell and some others.

I did do a markup language for screen-play-scripts for an animation engine for sprite actions and dialogue in the late 1990s. It's likely some some game engine frameworks support it, so it can be already done as an app for iOS and Android, but not as a standalone ebook.

[tomsem]

There is, in fact a spec:

http://idpf.org/epub/31/spec/epub-overview.html#sec-tts

[DNSB]

You might also want to check the EPUB 3.2 spec at 4.3.3 Text-to-Speech. As far as I can tell, there is no provision for multiple voices and the key word in that chunk of the spec is SHOULD which is generally translated as it's optional and no one is likely to implement it.

[Quoth]

Using epub (of any type) for the intention to support multiple TTS voices is crazy, because even if it works (doubtful) it likely needs a phone or tablet with iOS or Android ANYWAY, so use an App framework. Not sure if Unreal Engine can do it.

[murraypaul]

You'd really want a proper semantic language such as TEI, which does support this (https://tei-c.org/release/doc/tei-p5...f-speaker.html), rather than a layout language.

I did implement something like this as a toy version many years ago.

[Quoth]

Originally Posted by murraypaul

You'd really want a proper semantic language … rather than a layout language.

Absolutely. It's also not trivial to get right when there are bits with untagged speech. I've discovered ambiguities in texts when narrating an audio book. Really you need to at least record a narration to ensure you tag the dialogue properly, but it could be done using custom span class per speaker in HTML and then the CSS has speaker commented out for a regular renderer and your custom app then ignore the comment symbol.