Summary

CSS Speech properties offer the promise of improved aural rendering.

Examples

<abbr class="spell">IBM</abbr>
<span class="spell">IOU</span>

/* css */

.spell {
   speak-as: spell-out
}

<span class="digits">911</span>
<span class="digits">416 555-0123</span>
<span class="digits">90210</span>

/* css */

.digits {
   speak-as: digits
}

<p>
   Example one is correctly punctuated as follows:
   <span class="punctuate answer">The Franks, 
   a warlike people of Germany, gave their 
   name to France.</span>
</p>

/* css */

.punctuate {
   speak-as: literal-punctuation
}

<p>The telegram from Dr. King to President Kennedy
   read as follows:</p>
<blockquote>
   <pre class="silent">
   HOWEVER I AM CONVINCED THAT UNLESS SOME STEPS ARE TAKEN BY
   THE FEDERAL GOVERMENT TO RESTORE A SENSE OF CONFIDENCE IN
   THE PROTECTION OF LIFE, LIMB AND PROPERTY MY PLEAS SHALL FALL
   ON DEAF EARS AND WE SHALL SEE THE WORST RACIAL HOLOCAUST THIS
   NATION HAS EVER SEEN AFTER TODAYS TRAGEDY, INVESTIGATION WILL
   NOT SUFFICE.
   </pre>
   <cite><a 
	  href="http://www.jfklibrary.org/Asset-Viewer/-crU2bLgN0CcGkys8dkuHg.aspx"
	  >September 15, 1963 Telegram</a></cite>
</blockquote>

/* css */

.silent {
   speak-as: no-punctuation
}

h1 {
   pause: 50ms 25ms;
   cue: url('audio/ping.mp3') none;
   rest: 10ms 0ms
}

.male {
   voice-family: male
}

.female {
   voice-family: female
}

/* html */

<p class="female">
   Alice: But I don't want to go among mad people.
</p>

<p class="male">
   The Cat: Oh, you can't help that.
   We're all mad here. I'm mad. You're mad.
</p>

.huck {
   voice-family: child male 1
}

.tom {
   voice-family: child male 2
}

/* html */

<p class="tom">
   "Well—I—I—well, that ought to settle it, of course; 
   but I can't somehow seem to understand it no way.  
   Looky here, warn't you ever murdered AT ALL?"
</p>

<p class="huck">
   "No. I warn't ever murdered at all—I played it on 
   them. You come in here and feel of me if you don't 
   believe me."
</p>

Explanation

The CSS Speech module provides additional text-to-speech (TTS) enhancement functionality. Unlike PLS lexicons and SSML markup, the Speech module properties are not focused on defining the correct pronunciation of words.

The primary property the CSS Speech module adds for enhancing TTS playback is speak-as. This property provides the ability to control whether the TTS engine will read each character (setting to spell-out) or number (digit) in a string out. (See Example 1 and Example 2.) TTS engines often use unreliable tests based on the apparent wordiness of acronyms to determine whether to voice them, but this property allows you to override that behavior.

The speak-as property also takes the complimentary values literal-punctuation and no-punctuation. The values, as expected, control whether the TTS engine will voice punctuation.

The module also includes the speak property, which provides the ability to control TTS rendering of content, regardless of whether the containing element is visible or not. Setting the none value disables rendering on an element, and setting the normal value enables.

The following properties are focused on non-prosodic aspects of TTS playback.

pause

The pause property controls the amount of pause that occurs before and after the element that it is applied to. Pauses are typically used to identify transitions between major structures, such as between paragraphs and when new sections are beginning. TTS engines use punctuation to provide pauses within the flow of the narrative.

The value of the pause property is a time value indicating the pause length. If only a single value is specified:

pause: 50ms

that time is applied both before and after the associated element.

You can individually control the time to pause before and after by including a second time value:

pause: 50ms 0ms

The amount of pause specified occurs before any aural cue and rest at the start of the associated element, and after any rest and cue at the end of the element.

cue

The cue property provides the ability to uniquely identify elements with an aural sound. Cues are helpful in distinguishing new headings, for example, as pauses alone are not a good indicator.

Note that the cue property will render the associated audio clip both before and after the heading if only a single value is specified:

cue: url('audio/ping.mp3');

Users typically only expect a cue to signal the start, so use the null value to disable cues after the associated element has been rendered:

cue: url('audio/ping.mp3') null;

The aural cue occurs between any pause and rest at the start of the associated element, and between any rest and cue at the end of the element.

rest

The rest property controls the pause that occurs between the any aural cues and the rendering of the associated element, both before and after.

The value of the rest property is a time value indicating the pause length. If only a single value is specified:

rest: 25ms

that time is applied both before and after the associated element.

You can individually control the time to pause before and after by including a second time value:

rest: 25ms 0ms

The amount of rest specified occurs after any pause and cue at the start of the associated element, and before any cue and pause at the end of the element.

voice-family

The voice-family property provides control over the gender and type of voice used for TTS playback, allowing content producers to create more realistic TTS playback (e.g., alternating gender to match the character).

Although it's possible to name the voice to use:

voice-family: 'Dave';

in practice, with the wide variety of devices an EPUB may be played on, such specificity is only so useful as it requires knowing the names of all voices available on all devices.

Instead, it is better to request a voice using the pattern: age?, gender, integer? (where the question mark indicates the field is optional):

.king-lear {
   voice-family: old male 1;
}

The age value may be child, young or old; the gender male, female or neutral; and, when specified, the integer indicates the ordinal position of the voice to use (i.e., when more than one matching voice is available).

Frequently Asked Questions

Do I need to use the -epub- prefix on the properties?

No. As the properties were first introduced before the specification was stable, they were required to use a prefix. The unprefixed versions are now valid to use.

Can I force the TTS engine to say acronyms instead of spell them?

The Speech module does not provide a way to tell an engine it must voice a capitalized term. When including an acronym like EPUB, you would have to use a lexicon or attach an SSML pronunciation attribute to absolutely ensure that it does not get spelled out.

Why do I need to control the voicing of punctuation?

Although most engines will voice significant pause points, such as colons, they will typically not render each punctuation point in a document as it would ruin the reading experience. There are times when it is critical to ensure that the user is able to hear all the punctuation in a sentence or phrase, such as in grammar textbooks, programming guides and the like. (See Example 3.)

Accessible technologies also enable the pronunciation of all punctuation by default in elements such as pre and code. Although the benefit of reading all punctuation in computer code should be obvious, it is not always the case that preformatted text needs to such detailed rendering. Applying no-punctuation to a pre block of text ensures that it will be read without punctuation being announced.

Related Links

• EPUB 3 TTS Enhancements — CSS speech
• CSS — Speech Module