This document describes authoring features and reading system support for improving the voicing of EPUB®
3 publications.
Status of This Document
This section describes the status of this
document at the time of its publication. A list of current W3C
publications and the latest revision of this technical report can be found
in the W3C technical reports index at
https://www.w3.org/TR/.
This is a draft document and may be updated, replaced or obsoleted by other
documents at any time. It is inappropriate to cite this document as other
than work in progress.
The
W3C Patent
Policy
does not carry any licensing requirements or commitments on this
document.
The need for clear and accurate Text-to-Speech (TTS) rendering of publications is imperative for
their readability and comprehension. Unfortunately, the complexities of voicing natural languages
and the limitations of built-in vocabularies in TTS engines often leads to incorrect and illegible
voicing. Users either have to infer the correct meaning, when possible, or stop reading and have the
garbled words spelled out. Anyone who has tried to read educational or instructional material using
basic TTS playback will understand the frustration of this experience.
W3C has defined a variety of technologies to aid in improving the voice rendering of markup content:
the Synthetic Speech Markup Language [ssml], pronunciation lexicons [pronunciation-lexicon], and
the CSS Speech module.
SSML and pronunciation lexicons provide enhanced speech rendering. Lexicons are like dictionaries of
common terms a TTS engine can use, while SSML provides the ability to add individual voicing for
specific phrases. EPUB creators can use these technologies together or separately depending on
the complexity of the text. Despite these advantages, the technologies have not been adapted for
easy use within the XHTML and SVG formats that EPUB relies on. This document proposes an approach to
enable their authoring and rendering in EPUB content documents.
This document also covers the use of CSS Speech for improved aural rendering in EPUB. CSS Speech
covers a different domain than SSML and pronunciation lexicons. Instead of controlling the specific
voicing of words and phrases, these properties allow EPUB creators to aspects of the aural playback
itself — what text to render, at what volume, with what preferred voice, etc.
This document covers the use of these technologies for rendering by EPUB reading systems.
Although it is anticipated that general assistive technologies such as screen readers could take
advantage of the technologies, use by them is out of scope.
1.2 Background
This section is non-normative.
The EPUB Working Group of the International Digital Publishing Forum (IDPF) first defined a means of
integrating the Synthetic Speech Markup Language [ssml] and pronunciation lexicons
[pronunciation-lexicon] in EPUB 3.0 [epubcontentdocs-30] so that EPUB creators could improve
the rendering quality of text-to-speech (TTS) playback in reading systems. The ability to
include cascading style sheets [css2] also allowed EPUB creators to access the in-development
speech properties of the CSS Speech module [css-speech-1].
Although there has been some authoring uptake of these technologies, support in reading systems has
yet to materialize to a level where these technologies are considered stable. Consequently, these
technologies are now published as a W3C Working Group Note.
EPUB creators can continue to use these technologies in their publications, as the move to a Note
does not change their validity or affect backward compatibility. Developers of reading systems that
support TTS playback are also strongly encouraged to implement support. The Working Group will look
at standardizing any of the technologies that meet support requirements in future revisions of EPUB
3.
Note
The Specification for Spoken Presentation in HTML [spoken-html]
is another initiative in W3C to bring SSML to HTML. It is still too early to determine what
effect, if any, it will have on this document. The Working Group will monitor the work and
future updates to this Note will reflect any impact it has on Text-to-Speech rendering in
EPUB.
The rendering of the textual content of an EPUB publication by a reading system as
artificial human speech using a synthesized voice.
Note
Only the first instance of a term in a section links to its definition.
1.4 Conformance
As well as sections marked as non-normative, all authoring guidelines, diagrams, examples, and notes in this specification are non-normative. Everything else in this specification is normative.
The key words MAY, MUST, MUST NOT, SHOULD, and SHOULD NOT in this document
are to be interpreted as described in
BCP 14
[RFC2119] [RFC8174]
when, and only when, they appear in all capitals, as shown here.
2. SSML attributes
2.1 Introduction
This section is non-normative.
The W3C Speech Synthesis Markup Language [ssml] is a language used for assisting Text-to-Speech
(TTS) engines in generating synthetic speech. Although SSML is designed as a standalone document
type, it also defines semantics suitable for use within other markup languages.
The attributes allow EPUB creators to specify the proper phonetic pronunciation for uncommon terms
that a TTS engine is likely to mispronounce, as well as to disambiguate heteronyms.
2.2 The ssml:ph attribute
The ssml:ph attribute specifies a phonemic/phonetic pronunciation of the text
represented by its carrying element.
Attribute Name
ph
Namespace
https://www.w3.org/2001/10/synthesis
Usage
EPUB creators MAY specify on any element in EPUB content documents with which they can
logically associate a phonetic equivalent (i.e., that has descendant text content that a
Text-to-Speech engine would otherwise render).
EPUB creators MUST NOT specify the attribute on a descendant of an element that already
carries this attribute.
The ssml:ph attribute inherits the authoring requirements of the [ssml]
phoneme element's ph
attribute.
When the ssml:ph attribute appears on an element that has text node
descendants, the corresponding document text to which the pronunciation applies is the string that
results from concatenating the descendant text nodes, in document order. The specified phonetic
pronunciation must therefore logically match the element's textual data in its entirety (i.e., not
just an isolated part of its content).
EPUB creators SHOULD NOT use the ssml:ph attribute on elements without
text content that a Text-to-Speech engine would normally render (e.g., on empty div or
span elements). The attribute is not intended to add additional voicing only for
TTS playback, and reading systems are expected to ignore the attribute if it does not replace text
they would normally render.
Note
The ssml:ph attribute does not replace attribute values that carry additional
textual information (e.g., alt [html] and aria-label [wai-aria])
or link additional textual information (e.g., aria-describedby
[wai-aria]).
Similarly, EPUB creators SHOULD NOT add empty ssml:ph attributes to
try and suppress the rendering of text. Reading systems are expected to ignore empty attributes.
(See the aria-hidden
attribute [wai-aria] for specifying that content is only for visual rendering.)
2.3 The ssml:alphabet attribute
The ssml:alphabet attribute specifies which phonemic/phonetic pronunciation alphabet is
used in the value of the ssml:ph attribute.
Attribute Name
alphabet
Namespace
https://www.w3.org/2001/10/synthesis
Usage
EPUB creators MAY specify on any element in an EPUB content document that can contain
descendant text content.
Value
The name of the pronunciation alphabet used to express the value of the ssml:ph attribute.
The ssml:alphabet attribute inherits the authoring requirements of the [ssml]
phoneme element's alphabet attribute.
The value of the ssml:alphabet attribute is inherited in the
document tree. The pronunciation alphabet used for each ssml:ph attribute value is
determined by locating the first occurrence of the ssml:alphabet attribute starting
with the element on which the ssml:ph attribute appears, followed by the nearest
ancestor element.
EPUB creators SHOULD ensure that an alphabet is defined in scope for all phonemes expressed in ssml:ph attributes. Interoperability of playback
cannot be guaranteed in the absence of a declaration — reading systems may apply a default
alphabet, for example, or may not voice the phoneme.
Note
Although the [ssml] specification refers to a registry of alphabets, one has not been
published. As the charter of the W3C Voice Browser Working Group has expired, the Working Group
does not anticipate the publication of such a registry. EPUB creators therefore should reference
reading system support documentation to determine what alphabet values they support. Some common
alphabets include x-JEITA (also x-JEITA-IT-4002 and x-JEITA-IT-4006) and x-sampa.
3. Pronunciation lexicons
3.1 Introduction
This section is non-normative.
The W3C Pronunciation Lexicon Specification (PLS) [pronunciation-lexicon] defines syntax and
semantics for XML-based pronunciation lexicons to be used by Automatic Speech Recognition and
Text-to-Speech (TTS) engines.
Pronunciation lexicons allow EPUB creators to define a single global phonetic pronunciation that
reading systems can use for all instances of a term instead of having to tag every instance using
the SSML attributes. It is a much more efficient way of defining pronunciations for words with only
a single pronunciation, or where a particular pronunciation is predominant.
EPUB creators can use the [html] link element
and [svg] link
element to associate one or more lexicons with their respective EPUB content document
type. When reading systems process the documents, they can identify the linked lexicons and use them
to initiate text-to-speech playback.
3.2 Lexicon conformance
A pronunciation lexicon:
MUST meet the conformance constraints for XML documents defined in XML
Conformance [epub-33].
For both types of EPUB content document, the link element MUST have its rel
attribute set to "pronunciation" and its type attribute set to the media
type "application/pls+xml".
The CSS Speech [css-speech-1] module defines properties that allow EPUB creators to declaratively
control the aural rendering of EPUB content documents. It includes properties for specifying the
preferred Text-to-Speech voice, the volume level, and pauses and cues to perform when encountering
elements.
As EPUB content documents support the use of cascading style sheets [css2], EPUB creators MAY use CSS
Speech [css-speech-1] properties in their style sheet definitions.
5. reading system support
5.1 Introduction
Reading systems may implement Text-to-Speech playback in different ways depending on the type
of engine they use — one might only feed the text content of the document to the engine, for
example, while another could support full markup. This document tries to provide flexibility in its
requirements to allow for these differences. The only requirement is that the correct rendering
behavior result.
Although this document frames the enhancements in the context of a reading system with built-in
Text-to-Speech rendering capabilities, it is anticipated that any application or assistive
technology that can access the markup of an EPUB publication will be able to use these features to
provide improved voice rendering. Ensuring the technologies works with these applications is outside
the scope of this work, however.
It is not required that the reading system use a Text-to-Speech engine that supports
pronunciation lexicons so long as the lexemes are processed and applied correctly. A
reading system might, for example, transform the lexicon into an alternative dictionary
format its TTS engine supports.
SSML and Pronunciation Lexicons
Reading systems that support SSML and pronunciation lexicons:
MUST let any pronunciation instructions provided via the ssml:ph attribute take precedence in cases where a grapheme element
[pronunciation-lexicon] matches a text node of an element that carries the
ssml:ph attribute.
CSS Speech
This document adds no additional requirements for reading system support to those defined in
[css-speech-1].
A. Change log
This section is non-normative.
Note that this change log only identifies substantive changes since EPUB content documents 3.2
— those that affect conformance or are similarly noteworthy.
25-June-2021: Clarified processing of ssml:alphabet attribute and added additional
requirements for the ssml:ph attribute to avoid its use for adding or removing text
vocalization. See issue 1706.
25-June-2021: Clarified application of pronunciation lexicons. See issue 1705.
22-June-2021: Added that SSML and PLS can also be used with SVG content documents. See issue 1710.
B. Acknowledgements
This section is non-normative.
The following members of the EPUB 3 Working Group contributed to the development of this specification:
Shadi Abou-Zahra (Amazon)
Will AWAD (Newgen Knowledgeworks)
Noel Ray Barron (Apple Inc.)
Sofia Bautista (Legible Media Inc.)
Leah Brochu (National Network for Equitable Library Service)
Matthew C. Chan (House of Anansi Press)
Yu-Wei Chang (Taiwan Digital Publishing Forum)
Juan Corona (Legible Media Inc.)
Dave Cramer (W3C Invited Expert, chair)
Romain Deltour (DAISY Consortium)
Marisa DeMeglio (DAISY Consortium)
Brady Duga (Google LLC)
Reinaldo Ferraz (NIC.br - Brazilian Network Information Center)
Symon Flaming (Rakuten Group, Inc.)
John Foliot (W3C Invited Expert)
Teenya Franklin (Pearson plc)
Hadrien Gardeur (EDRLab)
Matt Garrish (DAISY Consortium)
Jen Goulden (Crawford Technologies)
David Hall (Apple Inc.)
Ivan Herman (W3C, staff contact)
Tetsu Hoshino (Kodansha, Publishers, Ltd.)
jasen huang (ByteDance)
Norikazu Ishizu (Kadokawa Corporation)
Norihito IYENAGA (Kodansha, Publishers, Ltd.)
Rick Johnson (VitalSource Technologies)
Ken Jones (Circular Software)
Antonio Kamiya (Dentsu Group Inc.)
Deborah Kaplan (W3C Invited Expert)
Bill Kasdorf (Book Industry Study Group)
Hiroshi Kawada (MEDIA DO Co., Ltd.)
George Kerscher (DAISY Consortium)
Kazuhito Kidachi (Mitsue-Links Co., Ltd.)
Masakazu Kitahara (Voyager Japan, Inc.)
Toshiaki Koike (Voyager Japan, Inc.)
Ryo Kuroda (ACCESS CO., LTD.)
Charles LaPierre (Benetech)
Dan Lazin (Google LLC)
Philippe Le Hegaret (W3C)
Laurent Le Meur (EDRLab)
YuYu Lin (Taiwan Digital Publishing Forum)
Farrah Little (National Network for Equitable Library Service)
Victor Lopes (Apple Inc.)
Karan Malhotra (Newgen Knowledgeworks)
Makoto Murata (DAISY Consortium)
Cristina Mussinelli (Fondazione LIA)
Yoichiro Nagao (Kodansha, Publishers, Ltd.)
Theresa O'Connor (Apple Inc.)
Yoshinori Ohmura (SHUEISHA Inc.)
Rachel Osolen (National Network for Equitable Library Service)