text2expression
v0.0.2
Published
Converts text into expressions that can be used to animate a vrm model
Downloads
1
Maintainers
Readme
text2expression
this repository can be useful for when you want to animate something based on spoken text but do not have access to the audio stream, such as when using the Web Speech Synthesis API
NPM
Heres the npm package link: https://www.npmjs.com/package/text2expression#text2expression
The text2expression function tries to estimate mouth positions of a VRM model based on a given text. It provides you with the estimated runtime of an audio file that may be created based on the text and an array of VRM Mouth Expressions, containing the values for the expressions of "a", "i", "u", "e" and "o" as well as the duration for which each expression should last.
Example usage:
import { text2expression } from 'text2expression-package';
async function convertTextToExpression() {
try {
const expressions = await text2expression('Hi, how are you today', 'en', 'path/to/your/ipaDict.txt');
console.log(ipaExpressions);
/*
Output will be:
{
duration: 1257.3, // in milliseconds
text: 'hajˈ . hawˈ ɑˈɹ juˈ tʌdejˈ',
all: [...], // VRMMouthExpression array
}
*/
} catch (error) {
console.error(error);
}
}
convertTextToExpression();
Expressions
The equivalent mouthexpressions for a given ipa character have been manually mapped.
The file ipa2VRMMouthExpression.ts
contains the full mapping.
This website served as a reference for the IPA: https://www.seeingspeech.ac.uk/ipa-charts/
This website served as a reference for the VRM Model: https://vrm-viewer-48655.web.app/
Estimations
Estimations are based on the female english browser Speech Synthesis API voice anyone can use from within chrome.
Each voice and language has different characteristics to consider, so each one has to have their own characteristics represented in the estimations made.
Despite the approximative nature, it works quite well.
IPA Dict
You can supply a path to an IPA Dictionary and specify your source language if you want. The dictionary should be a simple text file containing tab-separated {source-language} <=> ipa list like so:
absolute æˈbsʌluˌt absolutely æˌbsʌluˈtli absoluteness æˈbsʌluˌtnʌs ...
You can find some dictionaries here: https://github.com/open-dict-data/
be sure to remove the "/" at the start and the end of the ipa strings.
An english <-> IPA dictionary is included inside this repository under "informath/text-to-ipa/ipadict.txt"
CHANGELOG
0.0.2: Moved function exports out of type exports