Japanese text-to-speech engine binding for NodeJS
npm install node-openjtalk+ libiconv
MacOSX users do not need to install libiconv because OSX already includes it.
``sh`
$ npm install node-openjtalk
js
var fs = require('fs');
var Speaker = require('speaker');
var OpenJTalk = require('node-openjtalk').OpenJTalk;// pre-included HTS voice file
var fn_voice = OpenJTalk.voices.mei_normal;
// instantiate OpenJTalk with an HTS voice
var open_jtalk = new OpenJTalk({voice: fn_voice});
// synthesize a voice buffer from a text
open_jtalk.synthesize("すもももももももものうち", function(error, buffer) {
if (!error) {
// save as audio file
var fd = fs.openSync(__dirname + '/test.wav', 'w');
fs.write(fd, buffer, 0, buffer.length, null, function(err, written, buffer) {
fs.closeSync(fd);
});
}
});
// synthesize a voice synchronously
var buffer = open_jtalk.synthesizeSync("すもももももももものうち");
// flush to node-speaker
var speaker = new Speaker({
channels: 1,
bitDepth: 16,
sampleRate: 48000
});
speaker.end(buffer);
`API
$3
Instantiates OpenJTalk object.You must specify HTS voice file path with an option argument.
js
var OpenJTalk = require('node-openjtalk').OpenJTalk;
var open_jtalk = new OpenJTalk({voice: "/path/to/HTS voice file"});
`You can use pre-included voice files.
js
var OpenJTalk = require('node-openjtalk').OpenJTalk;
var open_jtalk = new OpenJTalk({voice: OpenJTalk.voices.m001});
`The pre-included voice files are:
+ m001
+ mei_angry
+ mei_bashful
+ mei_happy
+ mei_normal
+ mei_sad
$3
Synthesizes voice audio asynchronously from a text. The callback takes two arguments, error and buffer object.js
open_jtalk.synthesize("音声合成するテキスト", function(error, buffer) {
if (!error) {
// do something with buffer
}
});
`
The "optionHook" function lets you modify synthesis parameters. For example, following code makes speech speed double.js
open_jtalk.synthesize(text, function(option) {
return {
speed: option.speed * 2
};
}, function(error, buffer) {});
``The available parameters are:
+ samplingFrequency
+ fPeriod
+ alpha
+ beta
+ speed
+ addHalfTone
+ msdThreshold
+ gvWeightForSpectrum
+ gvWeightForLogF0
+ volume
+ audioBuffSize
Synthesizes voice audio synchronously from a text. This function returns buffer object.
node-openjtalk builds on OpenJTalk and hts_engine API, and is shiped with HTS Voice "NIT ATR503 M001" and "Mei". Thanks for their works.
Both OpenJTalk and hts_engine API are issued under the Modified BSD license.
HTS Voice "NIT ATR503 M001" and "Mei" are both issued under the Creative Commons Attribution 3.0 license.