MsEdgeTTS

[Update Dec 2025]
The Read Aloud API now requires a user agent matching the Microsoft Edge browser, and thus will not work in browsers other than Microsoft Edge. It will of course still work in any server-side runtime.

A simple Azure Speech Service module that uses the Microsoft Edge Read Aloud API.

Full support for SSML Only supports speak, voice, and prosody element types. The following is the default SSML object:

<speak version="1.0" xmlns="http://www.w3.org/2001/10/synthesis" xmlns:mstts="https://www.w3.org/2001/mstts"
       xml:lang="${this._voiceLang}">
    <voice name="${voiceName}">
        <prosody rate="${rate}" pitch="${pitch}" volume="${volume}">
            ${input}
        </prosody>
    </voice>
</speak>
Copy

Documentation on the SSML format can be found here . All supported audio formats can be found here.

Example usage

Make sure to escape/sanitize your user's input! Use a library like xml-escape.

Write to stream

import {MsEdgeTTS, OUTPUT_FORMAT} from "msedge-tts";

const tts = new MsEdgeTTS();
await tts.setMetadata("en-IE-ConnorNeural", OUTPUT_FORMAT.WEBM_24KHZ_16BIT_MONO_OPUS);
const {audioStream} = tts.toStream("Hi, how are you?");

audioStream.on("data", (data) => {
    console.log("DATA RECEIVED", data);
    // raw audio file data
});

audioStream.on("close", () => {
    console.log("STREAM CLOSED");
});
Copy

Connection and synthesis deadlines

const tts = new MsEdgeTTS({
    connectionTimeoutMs: 10000,
    synthesisTimeoutMs: 60000,
});
Copy

These are the defaults in milliseconds; set either to 0 to disable it. Values must be integers between 0 and 2147483647. The connection deadline bounds the WebSocket handshake, including any clock-correction retry, and rejects setMetadata() or a pending synthesis request if it expires. The synthesis deadline applies separately to each request: it starts when SSML is sent and resets when that request receives activity, so it does not limit the total audio duration. A synthesis timeout emits an error on its audio stream (handle it with audioStream.on("error", ...)), closes its metadata stream, and rejects toFile()/rawToFile(); other requests can continue on the same connection.

In Node.js, a handshake rejected with HTTP 403 and a valid server Date is retried once using the server's time for authentication. The correction is shared across instances and does not change the machine's clock. Browsers cannot expose these handshake headers, so automatic clock correction is unavailable there.

Write to file

import {MsEdgeTTS, OUTPUT_FORMAT} from "msedge-tts";

(async () => {
    const tts = new MsEdgeTTS();
    await tts.setMetadata("en-US-AriaNeural", OUTPUT_FORMAT.WEBM_24KHZ_16BIT_MONO_OPUS);
    const {audioFilePath} = await tts.toFile("./tmpfolder", "Hi, how are you?");  
})();
Copy

The directory passed to toFile() must already exist. The audio is always saved as audio.<extension> (for example audio.webm), plus metadata.json when sentence or word boundaries are enabled, so calling toFile() again with the same directory overwrites the previous files. To choose your own file name, use toStream() and pipe audioStream to a file instead.

Change voice rate, pitch and volume

import {MsEdgeTTS, OUTPUT_FORMAT} from "msedge-tts";

(async () => {
    const tts = new MsEdgeTTS();
    await tts.setMetadata("en-US-AriaNeural", OUTPUT_FORMAT.WEBM_24KHZ_16BIT_MONO_OPUS);
    const {audioStream} = await tts.toStream("Hi, how are you?", {rate: 0.5, pitch: "+200Hz"});
})();
Copy

Prosody options are passed per request as the second argument to toStream() or the third argument to toFile(). setMetadata() only accepts MetadataOptions, so rate, pitch or volume values passed to it are ignored.

Use an alternative HTTP Agent

Use a custom http.Agent implementation like https-proxy-agent or socks-proxy-agent.

import {MsEdgeTTS, OUTPUT_FORMAT} from "msedge-tts";
import {SocksProxyAgent} from 'socks-proxy-agent';

(async () => {
    const agent = new SocksProxyAgent("socks://your-name%40gmail.com:abcdef12345124@br41.nordvpn.com")
    const tts = new MsEdgeTTS(agent);
    await tts.setMetadata("en-US-AriaNeural", OUTPUT_FORMAT.WEBM_24KHZ_16BIT_MONO_OPUS);
    const {audioStream} = await tts.toStream("Hi, how are you?");
})();
Copy

Get sentence and word boundaries

import {MsEdgeTTS, OUTPUT_FORMAT} from "msedge-tts";

(async () => {
    const tts = new MsEdgeTTS();
    await tts.setMetadata("en-US-AriaNeural", OUTPUT_FORMAT.WEBM_24KHZ_16BIT_MONO_OPUS, {
        wordBoundaryEnabled: true,
        sentenceBoundaryEnabled: true
    });
    // as stream
    const {metadataStream} = await tts.toStream("Hi, how are you doing today hello hello hello?");
    /* ->
        {
          "Metadata": [
            {
              "Type": "SentenceBoundary",
              "Data": {
                "Offset": 1000000,
                "Duration": 35875000,
                "text": {
                  "Text": "Hi, how are you doing today hello hello hello?",
                  "Length": 46,
                  "BoundaryType": "SentenceBoundary"
                }
              }
            }
          ]
        }
     */
    // or as file
    const {metadataFilePath} = await tts.toFile("./tmpfolder", "Hi, how are you?");
    /* ->
       {
          "Metadata": [
            <all metadata combined>
          ]
        }
     */
})();
Copy

API

For the full documentation check out the API Documentation.

This library only supports promises.