HeadTTS

HeadTTS is a free JavaScript text-to-speech (TTS) solution that provides phoneme-level timestamps and Oculus visemes for lip-sync, in addition to audio output (WAV/PCM). It uses Kokoro neural model and voices, and inference can run entirely in the browser (via WebGPU or WASM), or alternatively on a Node.js WebSocket/RESTful server (CPU-based).

Pros: Free. Doesn't require a server in in-browser mode. WebGPU support. Uses neural voices with a StyleTTS 2 model. Great for lip-sync use cases and fully compatible with the TalkingHead. MIT licensed, doesn't use eSpeak NG or any other GPL-licensed module.
Cons: WebGPU is only supported by default in Chrome and Edge desktop browsers. Takes time to load on first load and uses a lot of memory. WebGPU support in onnxruntime-node is still experimental and not released, so server-side inference is relatively slow. English is currently the only supported language.

If you're using a Chrome or Edge desktop browser, check out the In-browser Demo!

The project uses websockets/ws (MIT License), hugginface/transformers.js (with ONNX Runtime) (Apache 2.0 License) and onnx-community/Kokoro-82M-v1.0-ONNX-timestamped (Apache 2.0 License) as runtime dependencies. For information on language modules and dictionaries, see Appendix B. Using jest for testing.

You can find the list of supported English voices and voice samples here.

In-browser Module: `headtts.mjs`

The HeadTTS JavaScript module enables in-browser text-to-speech using Module Web Workers and WebGPU/WASM inference. Alternatively, it can connect to and use the HeadTTS Node.js WebSocket/RESTful server.

Create a new HeadTTS class instance:

import { HeadTTS } from "./modules/headtts.mjs";

const headtts = new HeadTTS({
  endpoints: ["ws://127.0.0.1:8882", "webgpu"], // Endpoints in order of priority
  languages: ['en-us'], // Language modules to pre-load (in-browser)
  voices: ["af_bella", "am_fenrir"] // Voices to pre-load (in-browser)
});

CLICK HERE to see all the OPTIONS.

Option	Description	Default value
`endpoints`	List of WebSocket/RESTful servers or backends `webgpu` or `wasm`, in order of priority. If one fails, the next is used.	`["webgpu",` `"wasm"]`
`audioCtx`	Audio context for creating audio buffers. If `null`, a new one is created.	`null`
`transformersModule`	URL of the `transformers.js` module to load.	`"https://cdn.jsdelivr.net/npm/` `@huggingface/transformers@3.4.2` `/dist/transformers.min.js"`
`model`	Kokoro text-to-speech ONNX model (timestamped) used for in-browser inference.	`"onnx-community/` `Kokoro-82M-v1.0-ONNX-timestamped"`
`dtypeWebgpu`	Data type precision for WebGPU inference: `"fp32"` (recommended), `"fp16"`, `"q8"`, `"q4"`, or `"q4f16"`.	`"fp32"`
`dtypeWasm`	Data type precision for WASM inference: `"fp32"`, `"fp16"`, `"q8"`, `"q4"`, or `"q4f16"`.	`"q4"`
`styleDim`	Style embedding dimension for inference.	`256`
`audioSampleRate`	Audio sample rate in Hz for inference.	`24000`
`frameRate`	Frame rate in FPS for inference.	`40`
`languages`	Language modules to be pre-loaded.	[`"en-us"`]
`dictionaryURL`	URL to language dictionaries. Set to `null` to disable dictionaries.	`"../dictionaries"`
`voiceURL`	URL for loading voices.	`"https://huggingface.co/` `onnx-community/` `Kokoro-82M-v1.0-ONNX/` `resolve/main/voices"`
`voices`	Voices to preload (e.g., `["af_bella", "am_fenrir"]`).	`[]`
`splitSentences`	Whether to split text into sentences.	`true`
`splitLength`	Maximum length (in characters) of each text chunk.	`500`
`deltaStart`	Adjustment (in ms) to viseme start times.	`-10`
`deltaEnd`	Adjustment (in ms) to viseme end times.	`10`
`defaultVoice`	Default voice to use.	`"af_bella"`
`defaultLanguage`	Default language to use.	`"en-us"`
`defaultSpeed`	Speaking speed. Range: 0.25–4.	`1`
`defaultAudioEncoding`	Default audio format: `"wav"` or `"pcm"` (PCM 16-bit LE).	`"wav"`
`trace`	Bitmask for debugging subsystems (`0`=none, `255`=all): Bit 0 (1): Connection Bit 1 (2): Messages Bit 2 (4): Events Bit 3 (8): G2P Bit 4 (16): Language modules	`0`

Note: Model related options apply only to in-browser inference. If inference is performed on a server, server-specific settings will apply instead.

Connect to the first supported/available endpoint:

try {
  await headtts.connect();
} catch(error) {
  console.error(error);
}

Make an onmessage event handler to handle response messages. In this example, we use TalkingHead instance head to play the incoming audio and lip-sync data:

// Speak and lipsync
headtts.onmessage = (message) => {
  if ( message.type === "audio" ) {
    try {
      head.speakAudio( message.data, {}, (word) => {
        console.log(word);
      });
    } catch(error) {
      console.log(error);
    }
  } else if ( message.type === "custom" ) {
    console.error("Received custom message, data=", message.data);
  } else if ( message.type === "error" ) {
    console.error("Received error message, error=", message.data.error);
  }
}

CLICK HERE to see all the available class EVENTS.

Event handler	Description
`onstart`	Triggered when the first message is added and all message queues were previously empty.
`onmessage`	Handles incoming messages of type `audio`, `error` and `custom`. For details, see the API section.
`onend`	Triggered when all message queues become empty.
`onerror`	Handles system or class-level errors. If this handler is not set, such errors are thrown as exceptions. Note: Errors related to TTS conversion are sent to the `onmessage` handler (if defined) as messages of type `error`.

Setup the voice:

headtts.setup({
  voice: "af_bella",
  language: "en-us",
  speed: 1,
  audioEncoding: "wav"
});

The HeadTTS client is stateful, so you don't need to call setup again unless you want to change a setting. For example, if you want to increase the speed, simply call headtts.setup({ speed: 1.5 }).

Synthesize speech using the current voice setup:

headtts.synthesize({
  input: "Test sentence."
});

The above approach relies on onmessage event handler to receive and handle response messages and it is the recommended approach for real-time use cases. An alternative approach is to await for all the related audio messages:

try {
  const messages = await headtts.synthesize({
    input: "Some long text..."
  });
  console.log(messages); // [{type: 'audio', data: {…}, ref: 1}, {…}, ...]
} catch(error) {
  console.error(error);
}

The input property can be a string or, alternatively, an array of strings or inputs items.

CLICK HERE to see the available input ITEM TYPES.

Type	Description	Example
`text`	Speak the text in `value`. This is equivalent to giving a pure string input.	{ type: "text", value: "This is an example." }
`speech`	Speak the text in `value` with corresponding subtitles in `subtitles` (optional). This type allows the spoken words to be different that the subtitles.	{ type: "speech", value: "One two three", subtitles: "123" }
`phonetic`	Speak the model specific phonetic alphabets in `value` with corresponding `subtitles` (optional).	{ type: "phonetic", value: "mˈɜɹʧəndˌIz", subtitles: "merchandise" }
`characters`	Speak the `value` character-by-character with corresponding `subtitles` (optional). Supports also numbers that are read digit-by-digit.	{ type: "characters", value: "ABC-123-8", subtitles: "ABC-123-8" }
`number`	Speak the number in `value` with corresponding `subtitles` (optional). The number should presented as a string.	{ type: "number", value: "123.5", subtitles: "123.5" }
`date`	Speak the date in `value` with corresponding `subtitles` (optional). The date is presented as milliseconds from epoch.	{ type: "date", value: Date.now(), subtitles: "02/05/2025" }
`time`	Speak the time in `value` with corresponding `subtitles` (optional). The time is presented as milliseconds from epoch.	{ type: "time", value: Date.now(), subtitles: "6:45 PM" }
`break`	The length of the break in milliseconds in `value` with corresponding `subtitles` (optional).	{ type: "break", value: 2000, subtitles: "..." }

TODO: Add support for audio type.

An example using an array of input items:

{
  type: "synthesize",
  id: 14, // Unique request identifier.
  data: {
    input: [
      "There were ",
      { type: "speech", value: "over two hundred ", subtitles: ">200 " },
      "items of",
      { type: "phonetic", value: "mˈɜɹʧəndˌIz ", subtitles: "merchandise " },
      "on sale."
    ]
  }
}

You can add a custom message to the message queue using the custom method:

headtts.custom({
  emoji: "😀"
});

Custom messages can be used, for example, to synchronize speech with animations, emojis, facial expressions, poses, and/or gestures. You need to implement the custom functionality yourself within the message handler.

CLICK HERE to see all the class METHODS.

Method	Description
`connect( settings=null, onprogress=null, onerror=null )`	Connects to the specified set of `endpoints` set in constructor or within the optinal `settings` object. If the `settings` parameter is provided, it forces a reconnection. The `onprogress` callback handles `ProgressEvent` events, while the `onerror` callback handles system-level error events. Returns a promise. Note: When connecting to a RESTful server, the method sends a hello message and considers the connection established only if a text response starting with `HeadTTS` is received.
`clear()`	Clears all work queues and resolves all promises.
`setup( data, onerror=null )`	Adds a new setup request to the work queue. See the API section for the supported `data` properties. Returns a promise.
`synthesize( data, onmessage=null, onerror=null )`	Adds a new synthesize request to the work queue. If event handlers are provided, they override other handlers. Returns a promise that resolves with a sorted array of related messages of type `"audio"` or `"error"`.
`custom( data, onmessage=null, onerror=null )`	Adds a new custom message to the work queue. If event handlers are provided, they override other handlers. Returns a promise that resolves with the related message of the type `"custom"`.

NodeJS WebSocket/RESTful Server: `headtts-node.mjs`

Install (requires Node.js v20+):

git clone https://github.com/met4citizen/HeadTTS
cd HeadTTS
npm install

Start the server:

npm start

CLICK HERE to see the COMMAND LINE OPTIONS.

Option	Description	Default
`--config [file]`	JSON configuration file name.	`./headtts-node.json`
`--trace [0-255]`	Bitmask for debugging subsystems (`0`=none, `255`=all): Bit 0 (1): Connection Bit 1 (2): Messages Bit 2 (4): Events Bit 3 (8): G2P Bit 4 (16): Language modules	`0`

An example:

node ./modules/headtts-node.mjs --trace 16

CLICK HERE to see the configurable PROPERTIES.

Property	Description	Default
`server.port`	The port number the server listens on.	`8882`
`server.certFile`	Path to the certificate file.	`null`
`server.keyFile`	Path to the certificate key file.	`null`
`server.websocket`	Enable the WebSocket server.	`true`
`server.rest`	Enable the RESTful API server.	`true`
`server.connectionTimeout`	Timeout duration for idle connections in milliseconds.	`20000`
`server.corsOrigin`	Value for the `Access-Control-Allow-Origin` header. If `null`, CORS will not be enabled.	`*`
`tts.threads`	Number of text-to-speech worker threads, ranging from 1 to the number of CPU cores.	`1`
`tts.transformersModule`	Name of the transformers.js module to use.	`"@huggingface/transformers"`
`tts.model`	The timestamped Kokoro TTS ONNX model.	`"onnx-community/` `Kokoro-82M-v1.0-ONNX-timestamped"`
`tts.dtype`	The data type precision used for inference. Available options: `"fp32"`, `"fp16"`, `"q8"`, `"q4"`, or `"q4f16"`.	`"fp32"`
`tts.device`	Computation backend to use. Currently, the only available option for Node.js server is `"cpu"`.	`"cpu"`
`tts.styleDim`	The embedding dimension for style.	`256`
`tts.audioSampleRate`	Audio sample rate in Hertz (Hz).	`24000`
`tts.frameRate`	Frame rate in frames per second (FPS).	`40`
`tts.languages`	A list of languages to preload.	[`"en-us"`]
`tts.dictionaryPath`	Path to the language modules. If `null`, dictionaries will not be used.	`"./dictionaries"`
`tts.voicePath`	Path to the voice files.	`"./voices"`
`tts.voices`	Array of voices to preload, e.g., `["af_bella","am_fenrir"]`.	`[]`
`tts.deltaStart`	Adjustment (in ms) to viseme start times.	`-10`
`tts.deltaEnd`	Adjustment (in ms) to viseme end times.	`10`
`tts.defaults.voice`	Default voice to use.	`"af_bella"`
`tts.defaults.language`	Default language to use. Supported options: `"en-us"`.	`"en-us"`
`tts.defaults.speed`	Speaking speed. Range: 0.25–4.	`1`
`tts.defaults.audioEncoding`	Default audio encoding format. Supported options are `"wav"` and `"pcm"` (PCM 16bit LE).	`"wav"`
`trace`	Bitmask for debugging subsystems (`0`=none, `255`=all): Bit 0 (1): Connection Bit 1 (2): Messages Bit 2 (4): Events Bit 3 (8): G2P Bit 4 (16): Language modules	`0`

Appendix A: Server API reference

WebSocket API

Every WebSocket request must have a unique identifier, id. The server uses a Web Worker thread pool, and because work is done in parallel, the order of responses may vary. Therefore, each response includes a ref property that identifies the original request, allowing the order to be restored if necessary. The JS client class handles this automatically.

Request: `setup`

{
  type: "setup",
  id: 12, // Unique request identifier.
  data: {
    voice: "af_bella", // Voice name (optional)
    language: "en-us", // Language (optional)
    speed: 1, // Speed (optional)
    audioEncoding: 'wav' // "wav" or "pcm" (PCM 16bit LE) (optional)
  }
}

Request: `synthesize`

{
  type: "synthesize",
  id: 13, // Unique request identifier.
  data: {
    input: "This is an example." // String or array of input items
  }
}

The response message for synthesize request is either error or audio.

Response: `error`

{
  type: "error",
  ref: 13, // Original request id
  data: {
    error: "Error loading voice 'af_bella'."
  }
}

Response: `audio`

Returns an audio object metadata that can be passed on the TalkingHead speakAudio method once the audio content itself has been added.

{
  type: "audio",
  ref: 13,
  data: {
    words: ["This ","is ","an ","example."],
    wtimes: [443, 678, 780, 864],
    wdurations: [192 ,91 ,52 ,868],
    visemes: ["TH", "I", "SS", "I", "SS", "aa", "nn", "SS", "aa", "PP", "PP", "E", "DD"],
    vtimes: [443, 494, 550, 678, 729, 780, 801, 989, 1126, 1175, 1225, 1275, 1354],
    vdurations: [61, 66, 85, 61, 40, 31, 31, 69, 59, 60, 60, 89, 378],
    audioEncoding: "wav"
  }
}

The actual audio content will be delivered after this message as binary data (see the next response message).

Response: Binary (ArrayBuffer)

Binary data as an ArrayBuffer related to the previous audio message. Depending on the set audio encoding, either a WAV file (wav) or a chunk of raw PCM 16bit LE samples (pcm).

RESTful API

RESTful server API is a more simple alternative for WebSocket API. The REST server is stateless, so voice parameters must be included for each POST message. If you are using the HeadTTS client class, it handles this internally.

POST `/v1/synthesize`

JSON	Description
`input`	Input to synthesize. String or an array of input items. For a string of text, maximum 500 characters.
`voice`	Voice name.
`language`	Language code.
`speed`	Speed of speech.
`audioEncoding`	Either "wav" for WAV file or "pcm" for raw PCM 16bit LE audio.

OK response:

JSON	Description
`audio`	AudioBuffer for `"wav"` audio encoding, ArrayBuffer of raw PCM 16bit LE samples for `"pcm"` audio encoding.
`words`	Array of words.
`wtimes`	Array of word starting times for `words` in milliseconds.
`wdurations`	Array of word durations for `words` in milliseconds.
`visemes`	Array of Oculus viseme IDs: `'aa'`, `'E'`, `'I'`, `'O'`, `'U'`, `'PP'`, `'SS'`, `'TH'`, `'CH'`, `'FF'`, `'kk'`, `'nn'`, `'RR'`, `'DD'`, `'sil'`.
`vtimes`	Array of viseme starting times for `visemes` in milliseconds.
`vdurations`	Array of viseme durations for `visemes` in milliseconds.
`audioEncoding`	Audio encoding: `"wav"` or `"pcm"`.

Error response:

JSON	Description
`error`	Error message string

Appendix B: Language modules and dictionaries

American English, `en-us`

The American English language module is based on the CMU Pronunciation Dictionary from Carnegie Mellon University, containing over 134,000 words and their pronunciations. The original dataset is provided under a simplified BSD license, allowing free use for any research or commercial purpose.

In the Kokoro TTS model, the American English language data was trained using the Misaki G2P engine (en). Therefore, the original ARPAbet phonemes in the CMU dictionary have been converted to IPA and then to Misaki-compatible phonemes by applying the following mapping:

ɚ → [ ɜ, ɹ ], ˈɝ → [ ˈɜ, ɹ ], ˌɝ → [ ˌɜ, ɹ ]
tʃ → [ ʧ ], dʒ → [ ʤ ]
eɪ → [ A ], ˈeɪ → [ ˈA ], ˌeɪ → [ ˌA ]
aɪ → [ I ], ˈaɪ → [ ˈI ], ˌaɪ → [ ˌI ]
aʊ → [ W ], ˈaʊ → [ ˈW ], ˌaʊ → [ ˌW ]
ɔɪ → [ Y ], ˈɔɪ → [ ˈY ], ˌɔɪ → [ ˌY ]
oʊ → [ O ], ˈoʊ → [ ˈO ], ˌoʊ → [ ˌO ]
əʊ → [ Q ], ˈəʊ → [ ˈQ ], ˌəʊ → [ ˌQ ]

Note: During the dataset conversion, some vowels were reduced to reflect casual speech since HeadTTS is primarily designed for conversational use.

The final dictionary is a plain text file with around 125,000 lines (2,8MB). Lines starting with ;;; are comments. Each other line represents one word and its pronunciations. The word and its different possible pronunciations are separated by a tab character \t. An example entry:

MERCHANDISE	mˈɜɹʧəndˌIz

Out-of-dictionary (OOD) words are converted using a rule-based algorithm based on NRL Report 7948, Automatic Translation of English Text to Phonetics by Means of Letter-to-Sound Rules (Elovitz et al., 1976). The report is available here.

Finnish, `fi`

Important

As of now, Finnish language is not supported by the Kokoro model. You can use the fi language code with the English voices, but the pronunciation will sound rather weird.

The phonemization of the Finnish language module is done by an in-built algorithm. The algorithm doesn't require a pronunciation dictionary, but it uses a compound word dictionary to get the secondary stress marks right for compound words.

The dictionary used for compound words is based on The Dictionary of Contemporary Finnish maintained by the Institute for the Languages of Finland. The original dataset contains more than 100,000 entries and is open-sourced under the CC BY 4.0 license.

The pre-processed compound word dictionary is a plain text file with around 50,000 entries in 10,000 lines (~350kB). Lines starting with ;;; are comments. Each other line represents the first part of a compound word and the first four letters of all possible next words, all separated by a tab character \t. An example entry:

ALUMIINI	FOLI	KATT	OKSI	PAPE	SEOS	VENE	VUOK

Appendix C: Latency

In-browser TTS on WebGPU is 3x times faster than real-time and approximately 10x faster than WASM. CPU inference on a Node.js server performs surprisingly well, but increasing the thread pool size worsens performance, so we need to wait for WebGPU support. Quantization makes no significant difference, so I recommend using 32-bit floating point precision (fp32) for the best audio quality unless memory consumption becomes a concern.

Unofficial latency results using my own latency test app:

TTS Engine	Setup	`FIL`^[1]	`FBL`^[2]	`RTF`^[3]
HeadTTS, in-browser	Chrome, WebGPU/fp32	9.4s	958ms	0.30
HeadTTS, in-browser	Edge, WebGPU/fp32	8.7s	913ms	0.28
HeadTTS, in-browser	Chrome, WASM/q4	88.4s	8752ms	2.87
HeadTTS, in-browser	Edge, WASM/q4	44.8s	4437ms	1.46
HeadTTS, server	WebSocket, CPU/fp32, 1 thread	6.8s	712ms	0.22
HeadTTS, server	WebSocket, CPU/fp32, 4 threads	6.0s	2341ms	0.20
HeadTTS, server	REST, CPU/fp32, 1 thread	7.0s	793ms	0.23
HeadTTS, server	REST, CPU/fp32, 4 threads	6.5s	2638ms	0.21
ElevenLabs	WebSocket	4.8s	977ms	0.20
ElevenLabs	REST	11.3s	1097ms	0.46
ElevenLabs	REST, Flash_v2_5	4.8s	581ms	0.22
Microsoft Azure TTS	Speech SDK, WebSocket	1.1s	274ms	0.04
Google TTS	REST	0.79s	67ms	0.03

^[1] Finish latency: Total time from sending text input to receiving the full audio.

^[2] First byte/part/sentence latency: Time from sending the text input to receiving the first playable byte/part/sentence of audio. Note: This measure is not comparable across all models, since some solutions use streaming, some not.

^[3] Real-time factor = Time to generate full audio / Duration of the full audio. If RTF < 1, synthesis is faster than real-time (i.e., good).

CLICK HERE here to see the TEST SETUP.

Test setup: Macbook Air M2 laptop, 8 cores, 16GB memory, macOS Sequoia 15.3.2, Metal2 GPU 10 cores, 300/50 Mbit/s internet connection. The latest Google Chrome/Edge desktop browsers.

All test cases use WAV or raw PCM 16bit LE format and the "List 1" of the Harvard Sentences:

The birch canoe slid on the smooth planks.
Glue the sheet to the dark blue background.
It's easy to tell the depth of a well.
These days a chicken leg is a rare dish.
Rice is often served in round bowls.
The juice of lemons makes fine punch.
The box was thrown beside the parked truck.
The hogs were fed chopped corn and garbage.
Four hours of steady work faced us.
A large size in stockings is hard to sell.

Name		Name	Last commit message	Last commit date
Latest commit History 31 Commits
avatars		avatars
dictionaries		dictionaries
modules		modules
tests		tests
voices		voices
.gitattributes		.gitattributes
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
headtts-node.json		headtts-node.json
index.html		index.html
jest.config.mjs		jest.config.mjs
logo.png		logo.png
package-lock.json		package-lock.json
package.json		package.json

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Repository files navigation

HeadTTS

In-browser Module: `headtts.mjs`

NodeJS WebSocket/RESTful Server: `headtts-node.mjs`

Appendix A: Server API reference

WebSocket API

Request: `setup`

Request: `synthesize`

Response: `error`

Response: `audio`

Response: Binary (ArrayBuffer)

RESTful API

POST `/v1/synthesize`

Appendix B: Language modules and dictionaries

American English, `en-us`

Finnish, `fi`

Appendix C: Latency

About

Uh oh!

Releases 6

Uh oh!

Languages

License

met4citizen/HeadTTS

Folders and files

Latest commit

History

Repository files navigation

HeadTTS

In-browser Module: headtts.mjs

NodeJS WebSocket/RESTful Server: headtts-node.mjs

Appendix A: Server API reference

WebSocket API

Request: setup

Request: synthesize

Response: error

Response: audio

Response: Binary (ArrayBuffer)

RESTful API

POST /v1/synthesize

Appendix B: Language modules and dictionaries

American English, en-us

Finnish, fi

Appendix C: Latency

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 6

Uh oh!

Languages

In-browser Module: `headtts.mjs`

NodeJS WebSocket/RESTful Server: `headtts-node.mjs`

Request: `setup`

Request: `synthesize`

Response: `error`

Response: `audio`

POST `/v1/synthesize`

American English, `en-us`

Finnish, `fi`