|
<!doctype html> |
|
<html lang="en"> |
|
<head> |
|
<meta name="viewport" content="width=device-width" /> |
|
<link rel="stylesheet" href="style.css" /> |
|
<meta charset="UTF-8"> |
|
<title>Match-TTS Onnx Benchmarks</title> |
|
</head> |
|
<body> |
|
<h1>Match-TTS Onnx Benchmarks</h1> |
|
|
|
|
|
<script src="https://cdn.jsdelivr.net/npm/onnxruntime-web/dist/ort.webgpu.min.js" ></script> |
|
<script type="module"> |
|
import { MatchaTTSRaw } from "./js-esm/matcha_tts_raw.js"; |
|
import { webWavPlay } from "./js-esm/web_wav_play.js"; |
|
import { arpa_to_ipa } from "./js-esm/arpa_to_ipa.js"; |
|
import { loadCmudict } from "./js-esm/cmudict_loader.js"; |
|
import { env,textToArpa} from "./js-esm/text_to_arpa.js"; |
|
|
|
env.allowLocalModels = true; |
|
env.localModelPath = "./models/"; |
|
env.backends.onnx.logLevel = "error"; |
|
|
|
let matcha_tts_raw |
|
let cmudict ={} |
|
let speaking = false |
|
let total_infer_time=0 |
|
let count_infer=0 |
|
let loaded_model_name |
|
let load_time |
|
async function main(model_name) { |
|
if (typeof model_name !== 'string') { |
|
model_name ="en001_ep6399_univ_simplify" |
|
} |
|
|
|
|
|
console.log(model_name) |
|
if (speaking){ |
|
console.log("speaking return") |
|
} |
|
|
|
speaking = true |
|
console.log("main called") |
|
if(!matcha_tts_raw){ |
|
const load_startTime = performance.now(); |
|
matcha_tts_raw = new MatchaTTSRaw() |
|
console.time("load model"); |
|
const model_path = `./models/matcha-tts/${model_name}.onnx` |
|
console.log(model_path) |
|
await matcha_tts_raw.load_model(model_path,{ executionProviders: ['webgpu','wasm'] }); |
|
|
|
console.timeEnd("load model"); |
|
|
|
load_time = (performance.now() - load_startTime)/1000 |
|
loaded_model_name = model_name |
|
|
|
let cmudictReady = loadCmudict(cmudict,'./dictionaries/cmudict-0.7b') |
|
await cmudictReady |
|
|
|
|
|
update_infer_bench1() |
|
}else{ |
|
console.log("session exist skip load model") |
|
} |
|
const startTime = performance.now(); |
|
const text = document.getElementById('textInput').value |
|
console.log("### textToArpa call") |
|
const arpa_text = await textToArpa(cmudict,text) |
|
console.log("### arpa returned") |
|
const ipa_text = arpa_to_ipa(arpa_text).replace(/\s/g, ""); |
|
|
|
|
|
const spks = 0 |
|
const speed = document.getElementById('speed').value |
|
const tempature = document.getElementById('temperature').value |
|
|
|
console.time("infer"); |
|
const result = await matcha_tts_raw.infer(ipa_text, tempature, speed,spks); |
|
|
|
if (result!=null){ |
|
console.timeEnd("infer"); |
|
const endTime = performance.now(); |
|
const infer_time = endTime-startTime |
|
total_infer_time+=infer_time |
|
count_infer += 1 |
|
update_infer_bench2() |
|
webWavPlay(result) |
|
|
|
|
|
} |
|
|
|
speaking = false |
|
} |
|
function update_infer_bench1(){ |
|
|
|
const text = `${loaded_model_name} load time ${load_time.toFixed(1)} sec`; |
|
document.getElementById('result1').innerText=text |
|
} |
|
|
|
function update_infer_bench2(){ |
|
const avg = (total_infer_time/count_infer)/1000 |
|
const text = `Infer Count ${count_infer} avg infer-time ${avg.toFixed(1)} sec`; |
|
document.getElementById('result2').innerText=text |
|
} |
|
function update_range(){ |
|
const value = document.getElementById('spks').value |
|
let formattedNumber = value.toString().padStart(3, '0'); |
|
document.getElementById('spks_label').textContent = formattedNumber |
|
} |
|
function update_range2(){ |
|
const value = document.getElementById('temperature').value |
|
|
|
document.getElementById('tempature_label').textContent = value |
|
} |
|
function update_range3(){ |
|
const value = document.getElementById('speed').value |
|
|
|
document.getElementById('speed_label').textContent = value |
|
} |
|
|
|
window.onload = async function(){ |
|
|
|
document.getElementById('myButton').onclick = main; |
|
|
|
document.getElementById('temperature').onchange = update_range2 |
|
document.getElementById('speed').onchange = update_range3 |
|
} |
|
function loadModel(model_name){ |
|
total_infer_time=0 |
|
count_infer=0 |
|
matcha_tts_raw=null |
|
main(model_name) |
|
} |
|
|
|
function create_button(label, model_name) { |
|
|
|
const button = document.createElement('button'); |
|
button.style ="margin:4px;" |
|
button.textContent = label; |
|
|
|
|
|
button.onclick = function() { |
|
loadModel(model_name); |
|
}; |
|
return button |
|
} |
|
|
|
|
|
document.getElementById('buttons').appendChild(create_button("ljspeech","ljspeech_sim")) |
|
document.getElementById('buttons').appendChild(create_button("ljspeech-quantized","ljspeech_sim_q8")) |
|
document.getElementById('buttons').appendChild(create_button("vctk","vctk_univ_simplify")) |
|
document.getElementById('buttons').appendChild(create_button("vctk-quantized","vctk_univ_simplify_q8")) |
|
document.getElementById('buttons').appendChild(create_button("en001","en001_ep6399_univ_simplify")) |
|
document.getElementById('buttons').appendChild(create_button("en001-quantized","en001_ep6399_univ_simplify_q8")) |
|
document.getElementById('buttons').appendChild(document.createElement('br')) |
|
document.getElementById('buttons').appendChild(create_button("en001-t2-step01","en001_6399_T2_step01")) |
|
document.getElementById('buttons').appendChild(create_button("en001-t2-step05","en001_6399_T2_step05")) |
|
document.getElementById('buttons').appendChild(create_button("en001-t2-step10","en001_6399_T2_step10")) |
|
|
|
document.getElementById('buttons').appendChild(document.createElement('br')) |
|
document.getElementById('buttons').appendChild(create_button("en001-univ-step01","en001_6399_univ_step01")) |
|
document.getElementById('buttons').appendChild(create_button("en001-univ-step05","en001_6399_univ_step05")) |
|
document.getElementById('buttons').appendChild(create_button("en001-univ-step10","en001_6399_univ_step10")) |
|
|
|
|
|
|
|
|
|
|
|
</script> |
|
<div id="result1">Click button to load a model</div> |
|
|
|
<div id="buttons"></div> |
|
<br> |
|
<div id="result2">en001-T2 and en001-univ are experimental</div> |
|
<br><br> |
|
<input type="text" id="textInput" value ="Hello Huggingface." placeholder="Enter some text here..."> |
|
|
|
<button id="myButton">Text To Speak</button><br> |
|
|
|
|
|
<label for ="temperature" style="width: 110px;display: inline-block;">Temperature</label> |
|
<input type="range" id="temperature" min="0" max="1.0" value="0.5" step="0.1"/> |
|
<label for ="temperature" id="tempature_label">0.5</label><br> |
|
|
|
<label for ="speed" style="width: 110px;display: inline-block;">Speed</label> |
|
<input type="range" id="speed" min="0.1" max="2.0" value="1.0" step="0.1"/> |
|
<label for ="speed" id="speed_label">1.0</label> |
|
<br> |
|
<br> |
|
|
|
<div>almost load time 15 sec,short text TTS time 2 sec(my 2070super-gpu)</div><br> |
|
<div>Quantized version is too slow and exist just for Github Page 100MB limitation so far</div><br> |
|
<div>Multispeaker(vctk) is little bit slow than singlespeaker.default timesteps is 5(smallest 1 is 300msec fast,but audio become low quality)</div> |
|
<br> |
|
<div id="footer"> |
|
<b>Credits</b><br> |
|
<a href="https://github.com/akjava/Matcha-TTS-Japanese" style="font-size: 9px" target="link">Matcha-TTS-Japanese</a> | |
|
<a href = "http://www.udialogue.org/download/cstr-vctk-corpus.html" style="font-size: 9px" target="link">CSTR VCTK Corpus</a> | |
|
<a href = "https://github.com/cmusphinx/cmudict" style="font-size: 9px" target="link">CMUDict</a> | |
|
<a href = "https://huggingface.co/docs/transformers.js/index" style="font-size: 9px" target="link">Transformer.js</a> | |
|
<a href = "https://huggingface.co/cisco-ai/mini-bart-g2p" style="font-size: 9px" target="link">mini-bart-g2p</a> | |
|
<a href = "https://onnxruntime.ai/docs/get-started/with-javascript/web.html" style="font-size: 9px" target="link">ONNXRuntime-Web</a> | |
|
<a href = "https://github.com/akjava/English-To-IPA-Collections" style="font-size: 9px" target="link">English-To-IPA-Collections</a> | |
|
<a href ="https://huggingface.co/papers/2309.03199" style="font-size: 9px" target="link">Matcha-TTS Paper</a> |
|
</div> |
|
|
|
|
|
|
|
</body> |
|
</html> |
|
|
