Awesome
Speech Rule Engine
Breaking Change: Please move to v3.0.2 or later. More info here.
NodeJS version of the ChromeVox speech rule engine. Forked from ChromeVox release 1.31.0
Speech rule engine (SRE) can translate XML expressions into speech strings according to rules that can be specified in a syntax using Xpath expressions. It was originally designed for translation of MathML and MathJax DOM elements for the ChromeVox screen reader. Besides the rules originally designed for the use in ChromeVox, it also has an implemententation of the full set of Mathspeak and Clearspeak rules, localisation into a number of languages and Braille output currently in Nemeth.
SRE contains a library for semantic interpretation to re-represents any mathematical expression in its own internal semantic format, overcoming the poor design of presentation MathML by fully disassembling and reconstructing an expression. For a better understanding of the representation have a look at its visualiser. The semantic trees can be used in their own XML format directly or used to enrich the input MathML expressions with semantic information and speech strings.
There are three ways of using SRE:
-
Node Module: Download via npm or yarn. This is the easiest way to use the speech rule engine via its API and is the preferred option if you just want to include it in your project.
-
Standalone Tool: Download via github and build. This is useful if you want to use the speech rule engine in batch mode or interactivley to add your own code. Or simply run it with
npx
, for example to get all SRE options anywhere without local installation run:
npx speech-rule-engine -h
- Browser Library: This gives you the option of loading SRE in a browser and use its full functionality on your webesites.
Node Module
Install as a node module using npm:
npm install speech-rule-engine
Or add it with yarn:
yarn add speech-rule-engine
Then import into a running node or a source file using require:
require('speech-rule-engine');
API
Current API functions are divided into three categories.
Methods that take a string containing a MathML expression:
Method | Return Value |
---|---|
toSpeech(mathml) | Speech string for the MathML. |
toSemantic(mathml) | XML representation of the semantic tree for the MathML. |
toJson(mathml) | The semantic tree in JSON. |
toDescription(mathml) | The array of auditory description objects of the MathML expression. |
toEnriched(mathml) | The semantically enriched MathML expression. |
Note that in asynchronous operation mode for these methods to work correctly,
it is necessary to ensure that the Engine is ready for processing. In other
words, you need to wait for the setup promise to resolve. See documentation of
engineReady
and setupEngine
below.
Methods that take an input filename and optionally an output filename:
Note that the file methods only work in Node!
These methods load the given file, process its content and return the result. In asynchronous mode they return a promise that resolves to the result. If the output filename is given the result will also be written to this file. Note, that this will overwrite the existing file.
Method | Return Value |
---|---|
file.toSpeech(input, output) | Speech string for the MathML. |
file.toSemantic(input, output) | XML representation of the semantic tree for the MathML. |
file.toJson(input, output) | The semantic tree in JSON. |
file.toDescription(input, output) | The array of auditory description objects of the MathML expression. |
file.toEnriched(input, output) | The semantically enriched MathML expression. |
Methods for computing textual representations of numbers
These methods take a non-negative integer (either as string or number) as input
an return the number as text in the currently active locale. For the vulgar
method a vulgar fraction can be provided as a string of two slash separated
numbers, e.g., "1/2"
.
Method | Return Value |
---|---|
number(input) | The number as text. |
ordinal(input) | The word ordinal. |
numericOrdinal(input) | The numeric ordinal. |
vulgar(input) | Word representation of a vulgar fraction. |
Methods for querying and controlling the engine behaviour:
Method | Return Value |
---|---|
version | Returns SRE's version number. |
engineReady() | Returns a promise that resolves as soon as the engine is ready for processing (i.e., all necessary rule files have been loaded and the engine is done updating). This is important in asynchronous settings. |
setupEngine(options) | Takes an options feature vector to parameterise the Speech Rule Engine. Returns a promise that resolves as soon as the engine is ready for processing. |
engineSetup() | Returns the current setup of the engine as an options feature vector. |
Methods for navigating math expressions:
For the following methods SRE maintains an internal state, hence they are only useful when running in browser or in a Node REPL. Therefore, they are not exposed via the command line interface.
Method | Return Value |
---|---|
walk(input) | Speech string for the MathML. |
move(keycode) | Speech string after the move. Keycodes are numerical strings representing cursor keys, space, enter, etc. |
For more information on keybindings for the walker see here this dedicated page.
Options
There are a number of options that allow you to parameterise the Speech Rule
Engine. They can be set with the setupEngine(options)
method, which takes an
options feature vector (an object of option/value pairs) to parameterise the
engine. The engine's setup can be queried with the engineSetup()
method that
returns feature vector representing its current setup. Some options are quite
internal to SRE and are therefore not exposed via the command line interface.
Options to control speech output
Option | Value |
---|---|
domain | Domain or subject area of speech rules (e.g., mathspeak, clearspeak). |
style | Style or preference setting of speech rules (e.g., brief). |
In case of clearspeak, multiple preferences can be chosen using : as separator. | |
locale | Language locale in 639-1. |
subiso | More fine grained specification of locale. E.g., for French fr, be, or ch |
markup | Set output markup for speech: none , ssml , sable , voicexml , acss , ssml_step |
modality | Set the modality SRE returns. E.g., speech , braille , prefix , summary |
Observe that not every domain (i.e., speech rule set) implements every
style. Similarly, not every speech rule set is implemneted in every locale. For
a more detailed overview of locale, domain, style
combinations, use the
--opt
switch on the command line interface.
Options for enriched MathML output
Enriched MathML output is markup that embeds the internal semantic structure SRE uses into a modified represnentation of the original MathML. To get an idea of the semantic tree, take a look at its visualisation.
Option | Value |
---|---|
speech | Depth to which generated speech is stored in attributes during semantic enrichment. Values are none , shallow , deep . Default is none . |
pprint | Boolean flag to switch on pretty printing of output. This works on any XML style output. Default is true . |
structure | If set, includes a structure attribute in the enriched MathML that summarises the structure of the semantic tree in form of an sexpression. |
Options for internal control of the engine
These other options give more fine grained control of SRE. They can be useful during development and when integrating SRE into a larger project. They are given in decreasing order of interestingness.
Option | Value |
---|---|
json | URL where to pull the json speech rule files from. |
xpath | URL where to pull an xpath library from. This is important for environments not supporting xpath, e.g., IE or Edge. |
rate | Base value for speech rate in ssml_step markup |
strict | Boolean flag indicating if only a directly matching rule should be used. I.e., no default rules are used in case a rule is not available for a particular domain, style, etc. Default is false . |
mode | The running mode for SRE: sync , async , http |
By default SRE in node is in async , in browser in http , and on CLI in sync mode. | |
custom | Provide a custom method for locale loading. See below for more informaton. |
Custom Loading Method
SRE loads its locales and rule sets via loading methods specific for the particular environment and mode. I.e., it loads json files from the file system in node or via XML HTTP requests in the browser. These methods can be customised via passing a new method to the engine via the feature vector. A loader method takes the locale string as input and return a promise that resolves to the string containing the JSON structure of the locale file once loading is successfully completed. In other words it should be of type
(locale: string) => Promise<string>
In node the method can be directly set by passing it to the setupEngine
method
via the feature vector. As an example the following method loads locales from a
local folder at /tmp/mymaps
.
sre.setupEngine({
custom: loc => {
let file = `/tmp/mymaps/${loc}.json`;
return new Promise((res, rej) => {
try {
res(fs.readFileSync(file));
console.log(`Loading of locale ${loc} succeeded`);
} catch (_) {
console.log(`Loading of locale ${loc} failed`);
rej('');
}
});
}
});
Standalone Tool
Install the library using npm
:
npm install speech-rule-engine
Or add it with yarn:
yarn add speech-rule-engine
Run on command line
SRE can be run on the command line by providing a set of processing options and either a list of input files or a inputting an XML expression manually.
./node_modules/.bin/sre [options] infile1 infile2 infile3 ...
For example running
./node_modules/.bin/sre -j -p PATH_TO_SRE_RESOURCES/samples/sample1.xml PATH_TO_SRE_RESOURCES/samples/sample2.xml
will return the semantic tree in JSON as well as the speech translation for the
expressions in the two sample files.
(Note, that -p
is the default option if no processing options are given).
If you have npx
installed simply use this to run it from within the speech-rule-engine
folder with
npx sre
This works equivalently with all options and file input.
SRE also enables direct input from command line using stdin
. For example,
running
npx sre -j -p
will wait for a complete XML expression to be input for translation. Similarly, shell piping is allowed:
npx sre -j -p < PATH_TO_SRE_RESOURCES/samples/sample1.xml
Note, that when providing the -o outfile
option output is saved into the given file.
However, when processing from file only the very last output is reliably saved, while when
processing via pipes or command line input all output is saved.
Command Line Options
The following is a list of command line options for the speech rule engine.
Short | Long | Meaning |
---|---|---|
-o | --output [name] | Output file [name]. |
If not given output is printed to stdout. | ||
-d | --domain [name] | Domain or subject area [name]. |
This refers to a particular subject type of speech rules or subject area rules are defined for (e.g., mathspeak, clearspeak). | ||
If no domain parameter is provided, default is used. | ||
-t | --style [name] | Speech style [name]. |
Selects a particular speech style (e.g., brief). | ||
If no style parameter is provided, style default is used. | ||
-c | --locale | Language locale in ISO 639-1. |
-k | --markup [name] | Generate speech output with markup tags. Currently supported SSML, VoiceXML, Sable, ACSS (as sexpressions for Emacsspeak) |
-p | --speech | Generate speech output (default). |
-a | --audit | Generate auditory descriptions (JSON format). |
-j | --json | Generate JSON of semantic tree. |
-x | --xml | Generate XML of semantic tree. |
-P | --pprint | When given output is pretty printed if possible. |
-m | --mathml | Generate enriched MathML. |
-g | --generate [depth] | Include generated speech in enriched MathML. Supported values: none, shallow, deep (default: none) |
-w | --structure | Include structure attribute in enriched MathML. |
-N | --number | Translate number to words. |
-O | --ordinal | Translate number to word ordinal. |
-S | --numeric | Translate number to numeric ordinal. |
-F | --vulgar | Translate vulgar fraction to words. Provide vulgar fraction as slash seperated numbers. |
-C | --subiso | Subcategory of the locale given with -c. |
-v | --verbose | Verbose mode. Print additional information, useful for debugging. |
-l | --log [name] | Log file [name]. Verbose output is redirected to this file. |
If not given verbose output is printed to stdout. | ||
-h | --help | Enumerates all command line options. |
--opt | Enumerates all available options for locale, modality, domain and style. | |
-V | --version | Outputs the version number |
Building from Source
Clone from github and install dependencies either by running:
npm install
Or install them manually. SRE depends on the following libraries to run:
commander
xmldom-sre
wicked-good-xpath
In addition SRE depends on a number of libraries for development, in particular
TypeScript and
webpack as well as a number of plugins and libraries
to ensure source code hygiene. These are too numerous to list here and are best
viewed in the devDependencies
section of the package.json
file.
Running npm install
will build the single JavaScript modules, webpack the
bundle file in lib/sre.js
as well as assemble the locale files in
lib/mathmaps
. For more details on the build process as well as how to use
different bundlers see the developers documentation
below.
Run interactively
You can work with SRE interactively in node or use it as a batch processor, by
loading JavaScript modules directly. For the generation of speech make sure to
set the SRE_JSON_PATH
environment variable to the folder where your locale
files reside. For example, set
export SRE_JSON_PATH=./lib/mathmaps
before in the shell before running node in the speech-rule-engine
directory.
You can then load the individual modules simply with node's require
:
let sre = require('./js/index.js');
will give you the full API in the sre variable. On the other hand,
semantic = require('./js/semantic_tree/semantic.js');
will let you work with the semantic tree module only.
Browser Library
SRE can be used as a browser ready library giving you the option of loading it
in a browser and use its full functionality on your webesites. Since SRE version
4, the same bundle file works both in node and in a browser. Simply include the
file sre.js
in your website in a script tag
<script src="[URL]/sre.js"></script>
The API will now be available in the SRE
namespace.
Configuration
In addition to programmatically configuring SRE using the setupEngine
method, you can also include a configuration element in a website, that can take
the same options as setupEngine
.
For example the configuration element
<script type="text/x-sre-config">
{
"json": "https://cdn.jsdelivr.net/gh/zorkow/speech-rule-engine@develop/mathmaps/",
"xpath": "https://cdn.jsdelivr.net/gh/google/wicked-good-xpath@master/dist/wgxpath.install.js",
"domain": "mathspeak",
"style": "sbrief"
}
</script>
will cause SRE to load JSON files from rawgit and for IE or Edge it will also load Google's
wicked good xpath library. In addition the speech rules are set to mathspeak
in super brief
style.
Make sure the configuration element comes before the script tag loading SRE in your website!
Custom Loading Methods
Providing a custom loading method for locale loading in browser mode is very
similar to its use in node. However, note that since we now want to define a
function, it can not simply be given in the JSON configuration element in the
x-sre-config
script tag. Instead we need to define the special SREfeature
variable in the header of the file. Again make sure this script tag comes
before the script tag loading SRE in your website!
Here is an example of a custom function to load locales from localhost:
<script>
var SREfeature = {
"custom": function(loc) {
let file = 'http://localhost/sre/lib/mathmaps/' + loc + '.json';
let httpRequest = new XMLHttpRequest();
return new Promise((res, rej) => {
httpRequest.onreadystatechange = function() {
if (httpRequest.readyState === 4) {
console.log('Using my custom loader');
let status = httpRequest.status;
if (status === 0 || (status >= 200 && status < 400)) {
res(httpRequest.responseText);
} else {
rej(status);
}
}
};
httpRequest.open('GET', file, true);
httpRequest.send();
});
}
}
</script>
Developers Documentation
Since v4 SRE is distributed with the transpiled .js
files via the npm
package. These can be directly included into your respective projects. To
control the location of the locale mappings, set the SRE_JSON_PATH
environment
variable to the folder where your locale files reside. For example, set
export SRE_JSON_PATH=./lib/mathmaps
For more controlled integration or development fork or clone the repository from github.
Building
By default the build process consists of three steps:
- Creating Locale files in 'lib/mathmaps'
- Transpiling typescript from sources in 'ts/' to 'js/'
- Webpacking the bundle in 'lib/sre.js'
Locales are created from the sources in mathmaps
by combining the topically
split .json
files into a single, minimized .json
file, one for each
language. Note, that for ease of development JSON minimization is done via an
intermediate step to generate .min
files, which is handles in the Makefile
and ensures that only newly altered files have to be minimized.
While the entire build is best cleaned with make clean
, this does not clean
the .min
files. These should be cleaned with make clean_min
.
Alternative Builds
MathJax Library
This is only relevant for MathJax distributions version 2.7.X. As of version 3.0
MathJax uses sre directly via its npm
release. The MathJax specific build can
be created using the mjConfig
goal for webpack
npx webpack
that generates a build specific for MathJax in mathjax_sre.js
.
SRE can then be configured locally on webpages as described above.
Other Entry Points
For particular projects it might be useful and sufficient to create bespoke bundles of SRE submodules. Depending on the bundler
Examples of entry points that provide separate API functionality to SRE submodules are
Module | Javascript entry point | Typescript entry point | Comments |
---|---|---|---|
semantic | js/semantic_tree/semantic.js | ts/semantic_tree/semantic.ts | API for semantic tree generation |
enrich | js/enrich_mathml/enrich.js | ts/enrich_mathml/enrich.ts | API for semantic enrichment of MathML expressions. |
mathjax | js/common/mathjax.js | ts/common/mathjax.ts | API for MathJax version 2.7.X |
Using Bundlers
By default SRE works with webpack. But you can use a number of other bundlers to
build. Note, that for all bundlers the JSON locale files need to be build as
usual. Depending on the version of SRE you are using some of the alternative
bundlers might already be in the devDependencies
.
WebPack
The default webpack of SRE generates the sre.js
library that can be used both
in node and the browser. Other goals are mjConfig
that is defined in
webpack.config.js
or alternative entry points as discussed in more detail in
the Alternative Builds section.
Rollupjs
Creates bundles similar to webpack, that can be used pretty much as
replacement. Rollup bundles are build from the Javascript modules in js
, so
make sure to transpile first with npx tsc
.
Build
Install the required packages:
npm install --no-save rollup
npm install --no-save @rollup/plugin-commonjs
npm install --no-save @rollup/plugin-node-resolve
npm install --no-save rollup-plugin-terser
Add a rollup.config.js
file of the form:
import resolve from '@rollup/plugin-node-resolve';
import commonjs from '@rollup/plugin-commonjs';
import { terser } from "rollup-plugin-terser";
import pkg from './package.json';
export default [
{
input: 'js/index.js',
output: {
name: 'SRE',
file: pkg.browser,
format: 'umd'
},
plugins: [
resolve(),
commonjs(),
terser()
]
}
];
The resulting file in ./lib/sre.js
works as usual for both browser and node.
Note, that in the Browser version there is currently an issue with Unicode Braille Spaces that are being garbled.
Esbuild
Creates small builds very fast, but has some backward compatibility issues, in that it can work with ES6 modules only.
Build
Install the required packages:
npm install --no-save esbuild
Build using the following command line:
npx esbuild ./ts/index.ts --bundle --outfile=lib/sre.js --format=esm --minify
Node
To run in node you need ES6 modules enabled.
npx install --no-save esm
Then start node with node -r esm
. If you want to use the CLI interface, adapt
the ./bin/sre
script as follows:
#!/usr/bin/env -S node -r esm
let sre = require('../lib/sre.js');
(new sre.cli()).commandLine();
Browser
Import the SRE library as ES6 module into the browser, e.g.,
<script type="module">
import * as SRE from 'http://localhost/sre/speech-rule-engine/lib/sre.js'
...
</script>
Parcel
SRE currently does not support bundling with parcel
.
Coding Style
SRE is implemented using coding format enforced by prettier and eslint. In addition it requires full documentation using JSDOC. When creating a pull request, please make sure that your code compiles and is fully linted.
Code Formatting
We use prettier for formatting code. Run
npm format
to see which files contain format violations. Use
npm format-fix
to automatically fix those format violations.
Code Hygiene
We use eslint for enforcing code style and documentation. Run
npm lint
to see a list of linting errors and warnings. Use
npm lint-fix
to automatically fix those violations as much as possible.
Documentation
Full JSDOC documentation is required and enforced via the
eslint jsdoc plugin
. To
generate documentation from the JSDOC, simply run
npm run docs
This will generate documentation using typedoc for the
source code in the directory docs/
.
Node Package
The speech rule engine is published as a node package in fully compiled form, together with the JSON libraries for translating atomic expressions. All relevant files are in the lib subdirectory.
To publish the node package run
npm publish
For manually going through the build steps see the prepublish
script in package.json
.
Removed or Deprecated Functionality
The following is an overview of functionality and options that were available at some point in SRE. Depending on the version you are using they might still work, do nothing or throw an error. Even if they still work they are strongly discouraged to use.
Removed or Deprecated Engine Setup Options
The following options are either deprecated or have been removed. Having them in
the feature vector for setupEngine
should not throw an exception but will have
no effect.
Option | Value | Release | Comments |
---|---|---|---|
cache | Boolean flag to switch expression caching during speech generation. Default is true . | Removed in v3.2.0 | Expression caching has been removed and the option has no longer any effect. |
rules | A list of rulesets to use by SRE. This allows to artificially restrict available speech rules, which can be useful for testing and during rule development. Always expects a list, even if only one rule set is supplied! | Deprecated in v4.0.0 | Note that setting rule sets is no longer useful with the new rule indexing structures. It is only retained for testing purposes. |
walker | A walker to use for interactive exploration: None , Syntax , Semantic , Table | Deprecated since v4.0.0 | Defaults to Table walker. Other walkers are no longer maintained! |
semantics | Boolean flag to switch OFF semantic interpretation. | Removed in v3.0 | Non-semantic rule sets have been removed since v3.0. |
Removed API functions
Method | Return Value | Release | Comments |
---|---|---|---|
pprintXML(string) | Returns pretty printed version of a serialised XML string. | Removed v3.0 | Use the pprint option instead.` |
Removed Command Line Options
Short | Long | Meaning | Release | Comments |
---|---|---|---|---|
-i | --input [name] | Input file [name]. | Deprecated since v3.0. Removed in v4.0! | Use standard input file handling or stdio piping instead |
-s | --semantics | Switch OFF semantics interpretation. | Removed in v3.0 | There is no longer support for non-semantic rule sets. |
Breaking Change
Due to a breaking change in
the commander library
old versions of SRE might break when newly installed from npm
. SRE relies on
commander for running its command line interface.
Therefore, please move to SRE v3.0.2 or later.
If you want to run older versions of SRE, you need to manually downgrade the
commander
package to v5.1.0
by running
npm install commander@5.1.0
Note, that the command line option --options
has now been renamed to --opt
.