Voice JavaScript SDK: Changelog

In version 2.5.0, the SDK introduced a mechanism to override WebRTC APIs, enabling support for redirection technologies like Citrix HDX. In this release, the SDK extends this capability by allowing the override of the native MediaStream API, making it possible to create custom media streams tailored for such environments. See WebRTC API Overrides for an example.

• Fixed an issue where device.destroy() did not work for Chrome Extension Manifest V3.
• Addressed an issue where the publisher was null upon destruction of an AudioProcessorObserver.

• Added an API for testing PreflightTest in realms other than production. You can now pass chunderw and eventgw values within the options object when constructing a PreflightTest. Note: These new options are intended for internal testing by Twilio employees only, and should not be used otherwise.

• Fixed an issue where the originalError property is not populated when a general ConnectionError (31005) happens.

• Fixed an issue where the chunderw parameter is not being used during signaling reconnection. Note that this parameter is intended solely for testing purposes.

• Fixed an issue where calling device.connect() without waiting for the promise to get resolved, then calling device.audio.setInputDevice() right away results in an AcquisitionFailedError.

The Call Message Events, originally released in 2.2.0, has been promoted to GA. This release includes the following breaking changes.

• Call.Message.messageType has been converted from Call.MessageType enum to string.
• Call Message related errors are now emitted via call.on('error', handler(twilioError)) instead of device.on('error', handler(twilioError)).
• A new error, 31210, has been added to the SDK. This new error is emitted via call.on('error', handler(twilioError)) after calling the sendMessage API with an invalid Call.Message.messageType.
• Fixed an issue where the wrong error code 31209 is raised if the payload size of a Call Message Event exceeds the authorized limit. With this release, 31212 is raised instead.

• Fixed an issue where PreflightTest throws an error when RTCIceCandidateStatsReport is not available. Thanks to @phi-line for your contribution.

• The SDK now updates its internal device list when microphone permissions change.

• Fixed an issue where an AcquisitionFailedError is raised when making a call while a setInputDevice invocation is still in progress. The following snippet will reproduce the issue in versions <2.11.2:

1
// Call setInputDevice without waiting for it to resolve e.g. using 'await'
2
device.audio.setInputDevice(id);
3

4
// Calling device.connect immediately raises an AcquisitionFailedError error
5
device.connect(...);

• Fixed an issue where the input stream stops working after changing the default input device. Thanks @varunm0503 for your contribution.
• Fixed an echo issue where the audio output is duplicated after the device permission is granted. Thanks @kmteras for your contribution.

In Manifest V2, Chrome Extensions had the capability to run the Voice JS SDK in the background for making calls. However, with the introduction of Manifest V3, running the Voice JS SDK in the background is now possible only through service workers. Service workers lack access to certain features, such as DOM, getUserMedia, and audio playback, making it impossible to make calls with previous versions of the SDK.

This new release enables the SDK to run in a service worker context, allowing it to listen for incoming calls or initiate outgoing calls. Once a call object is created, it can be forwarded to an offscreen document, where the SDK gains access to all necessary APIs to fully establish and interact with the call. For implementation details, refer to our example.

Previous versions of the SDK supported simultaneous outgoing and incoming calls using different identities. If an incoming call was received while the Device with the same identity was busy, the active call had to be disconnected before accepting the incoming call. With this SDK release, multiple incoming calls for the same identity can be accepted, muted, or put on hold, without disconnecting any existing active calls. This is achieved by forwarding the incoming call to a different Device instance. Below are the new APIs and an example for more details.

• Call.connectToken
• ConnectOptions.connectToken

• Fixed an issue where an error is thrown when rtcConstraints parameter is provided.
• Fixed an issue (#118, #210) where certain calls are not ended right away after a page refresh.

• Fixed an issue where device.register() does not return a promise rejection when the WebSocket fails to connect. Thank you @kamalbennani for your contribution.
• Fixed an issue where audio processor insights events are not generated if there is an existing processed stream at the start of a call.

• Added tags to client logs for easier filtering
• Added log statements to API calls and events for debugging purposes

• Fixed an issue where updating token after signaling connection has gone offline causes an Invalid State error.
• Fixed an issue where Device.Options.logLevel is only accepting a number type. With this release, strings are now also allowed. See Device.Options.logLevel for a list of possible values.
• Fixed an issue where call.mute() does not have an effect while the call.status() is either ringing or connecting. Thank you @zyzmoz for your contribution.

The SDK now includes Audio Processor APIs, enabling access to raw audio input and the ability to modify audio data before sending it to Twilio. With this new feature, the following use cases can now be achieved on the client side:

• Background noise removal using a noise cancellation library of your choice
• Music playback when putting the call on hold
• Audio filters
• AI audio classification
• ... and more!

Please visit this page for more details about the Audio Processor APIs.

Added a new feature flag enableImprovedSignalingErrorPrecision to enhance the precision of errors emitted by Device and Call objects.

1
const token = ...;
2
const device = new Device(token, {
3
    enableImprovedSignalingErrorPrecision: true,
4
});

The default value of this option is false.

When this flag is enabled, some errors that would have been described with a generic error code are now described with a more precise error code. With this feature, the following errors now have their own error codes. Please see this page for more details about each error.

Device Error Changes

1
const device = new Device(token, {
2
    enableImprovedSignalingErrorPrecision: true,
3
});
4
device.on('error', (deviceError) => {
5
    // the following table describes how deviceError will change with this feature flag
6
});

Call Error Changes

1
const device = new Device(token, {
2
    enableImprovedSignalingErrorPrecision: true,
3
});
4
const call = device.connect(...);
5
call.on('error', (callError) => {
6
    // the following table describes how callError will change with this feature flag
7
});

IMPORTANT: If your application logic currently relies on listening to the generic error code 53000 or 31005, and you opt into enabling the feature flag, then your application logic needs to be updated to anticipate the new error code when any of the above errors happen.

• Fixed an issue where, sometimes a TypeError is raised while handling an incoming call under the following circumstances:
  • Network interruptions
  • updating the token before accepting the call

Updated November 1, 2023

We have identified an issue on Chromium-based browsers running on MacOS 14 (Sonoma) where the audio deteriorates during a call. This issue happens due to the excessive calls to MediaDevices: enumerateDevices() API. With this release, the SDK calls this API only when necessary to avoid audio deterioration.

• Fixed an issue where audio in the Chrome browser is choppy when another application is also using the audio devices.
• Added missing documentation for the following events:
  • call.on('ringing', handler)
  • call.on('warning', handler)
  • call.on('warning-cleared', handler)
  • device.on('destroyed', handler)

• Fixed an issue where call.sendMessage() API throws an error if the SDK is imported as an ECMAScript Module (ESM) using the @twilio/voice-sdk/esm path.

Currently, the SDK is imported as a CommonJS Module (CJS) using the root path @twilio/voice-sdk. With this release, the SDK contains an experimental feature that allows it to be imported as an ECMAScript Module (ESM) using the @twilio/voice-sdk/esm path. As this is an experimental feature, some frameworks using bundlers like Vite and Rollup may not work. Full support for ESM will be available in a future release and will become the default import behavior of the SDK.

Example:

import { Device } from '@twilio/voice-sdk/esm';

• Fixed some security vulnerabilities shown by npm audit.
• Removed unused dependencies.
• Replaced deprecated dependencies.

• Fixed an issue where calling device.updateOptions would reset the device.audio._enabledSounds state.
• Fixed an issue where custom DTMF sounds would not play. With this release, custom DTMF sounds should now play when configured during device initialization.

1
const device = new Device(token, {
2
  sounds: {
3
    dtmf8: 'http://example.com/8_button.mp3',
4
    // Other custom sounds
5
  },
6
  // Other options
7
});

• The SDK now builds on NodeJS versions 16 and above without the --legacy-peer-deps flag.
• Removed usage of NodeJS modules from the SDK and some dependencies. With this change, the SDK should now work with some of the latest frameworks that use the latest versions of bundlers such as Vite and Webpack.
• The AudioPlayer dependency has been incorporated into the SDK as part of a migration. This change fixes an issue where source maps are not properly loaded.
• Removed unnecessary files from the generated npm package.
• Links to source maps are now included in the generated npm package.
• The ws package has been moved to devDependencies.
• The SDK no longer depends on the xmlhttprequest npm package.

Updated: This is now GA as of December 14, 2023

The SDK now allows you to override WebRTC APIs using the following options and events. If your environment supports WebRTC redirection, such as Citrix HDX's WebRTC redirection technologies, your application can use this new beta feature for improved audio quality in those environments.

• Device.Options.enumerateDevices
• Device.Options.getUserMedia
• Device.Options.RTCPeerConnection
• call.on('audio', handler(remoteAudio))

• Updated the description of Device.updateToken API. It is recommended to call this API after Device.tokenWillExpireEvent is emitted, and before or after a call to prevent a potential ~1s audio loss during the update process.
• Updated stats reporting to stop using deprecated RTCIceCandidateStats - ip and deleted.

• Fixed an issue where a TypeError is thrown after rejecting a call then invoking updateToken.
• Fixed an issue (#87, #145) where the PeerConnection object is not properly disposed.
• Fixed an issue where device.audio.disconnect, device.audio.incoming and device.audio.outgoing do not have the correct type definitions.
• Fixed an issue where the internal deviceinfochange event is being emitted indefinitely, causing high cpu usage.

• Fixed an issue where a Twilio error is not returned when microphone access is blocked. Thank you @SiimMardus and @ostap0207 for your contribution.

• Fixed an issue where incoming sound will not stop playing after the call is disconnected. Thank you @kamalbennani for your contribution.

This release includes updated DNS names for Twilio Edge Locations. The Voice JS SDK uses these Edge Locations to connect to Twilio's infrastructure via the parameter Device.Options.edge. The current usage of this parameter does not change as the SDK automatically maps the edge value to the new DNS names.

Additionally, you need to update your Content Security Policies (CSP) if you have it enabled for your application. You also need to update your network configuration such as firewalls, if necessary, to allow connections to the new DNS names and IP addresses.

The SDK can now send and receive custom messages to and from Twilio's backend via the following new Call APIs.

• sendMessage
• messageReceivedEvent
• messageSentEvent

Please visit this page for more details about this feature. Additionally, please see the following for more information on how to send and receive messages on the server.

• UserDefinedMessage
• UserDefinedMessageSubscription

NOTE: This feature should not be used with PII.

Example

1
const device = new Device(token, options);
2

3
const setupCallHandlers = call => {
4
  call.on('messageReceived', message => messageReceivedHandler(message));
5
  call.on('messageSent', message => messageSentHandler(message));
6
};
7

8
// For outgoing calls
9
const call = await device.connect();
10
setupCallHandlers(call);
11

12
// For incoming calls
13
device.on('incoming', call => setupCallHandlers(call));
14
await device.register();
15

16
// For sending a message
17
const eventSid = call.sendMessage({
18
  content: { foo: 'foo' },
19
  messageType: Call.MessageType.UserDefinedMessage,
20
});

• Fixed an issue where insights data stops getting published after calling device.updateOptions.

• Ignoring a call will now properly stop the ringing sound
• NPM versioning has been fixed to specify >=12 rather than exactly 12
• Use DOMException instead of DOMError, which has been deprecated
• Removed npm util from the package, instead favoring native functions

The Voice JavaScript SDK now fully supports Call reconnection. If the media connection or signaling websocket is lost, the SDK is able to attempt to reconnect the Call. A Call can now potentially be recovered up to 30 seconds after a media or signaling connection loss.

The Twilio.Device will emit a 'reconnecting' event when a connectivity loss occurs, and a 'reconnected' event upon successful reconnection.

There exists a limitation such that Signaling Reconnection and Edge Fallback are mutually exclusive. To opt-in to the Signaling Reconnection feature, a new option can be passed to the SDK: maxCallSignalingTimeoutMs. If this value is not present in the options object passed to the Device constructor, the default value will be 0. Reconnection can only happen with an up-to-date AccessToken.

The Voice JavaScript SDK now provides two additional features to help keep your AccessTokens up to date:

1. The 'tokenWillExpire' event, which will be emitted by the Twilio.Device before its associated AccessToken is set to expire. By default, it will be emitted 10 seconds before the AccessToken's expiration.
2. The DeviceOptions.tokenRefreshMs property that can configure the timing of the 'tokenWillExpire' event.

You can use these new features in conjunction with the device.updateToken() method to automatically keep an AccessToken up to date.

In the following example, the 'tokenWillExpire' event will be emitted 30 seconds (3000 milliseconds) before the AccessToken is set to expire, and the event listener for the 'tokenWillExpire' event will retrieve a new AccessToken and update the Device's AccessToken with the device.updateToken() method.

1
const device = new Device(token, {
2
  // 'tokenWillExpire' event will be emitted 30 seconds before the AccessToken expires
3
  tokenRefreshMs: 30000,
4
});
5

6
device.on('tokenWillExpire', () => {
7
  return getTokenViaAjax().then(token => dev.updateToken(token));
8
});

The Twilio Voice JavaScript SDK now supports Twilio Regions.

If you are part of the Twilio Regions Pilot and wish to specify a home region when using the Voice JavaScript SDK, you will need to:

1. Create AccessTokens with API Keys and API Key Secrets that are stored in the specified Twilio Region, and include the Region name when creating the AccessToken. See example below.
2. Use an edge location that matches your specified Region when instantiating your Twilio.Device.

Below is an example of how you would use the Node.js Helper Library to create AccessTokens for the Voice JavaScript SDK for a Region.

1
const accessToken = const accessToken = new twilio.jwt.AccessToken(
2
  credentials.accountSid,
3
  credentials.apiKeySid,
4
  credentials.apiKeySecret, {
5
    identity,
6
    ttl,
7
    region: 'au1',
8
  },
9
);
10

11
const grant = new VoiceGrant({
12
  outgoingApplicationSid: credentials.twimlAppSid,
13
  incomingAllow: true,
14
});
15

16
accessToken.addGrant(grant);

Note: The API Key and Secret above must be created within the au1 region. It's recommended that the TwiML App used in the Voice Grant is also created in the same Region.

The example below shows how to pass the au1-related edge location to the Twilio.Device constructor.

1
const device = new Device(accessToken, {
2
  edge: 'sydney',
3
});

The new Twilio.Device.home accessor will return a string value of the home region of the device instance, given that it successfully connected with Twilio.

The Voice JavaScript SDK now exposes a Twilio.Device.identity accessor.

Given that a Twilio.Device has registered successfully with Twilio, the Twilio.Device.identity accessor will return a read-only string containing the identity that was passed to the AccessToken used to instantiate the Twilio.Device.

• Updated ws version to fix a potential security vulnerability
• All event listeners will now be properly cleaned up after calling Twilio.Device.destroy()
• When Insights fails to post an event, the SDK now logs a warning rather than an Uncaught Promise Rejection

Device must now be instantiated before it can be used. Calling Device.setup() will no longer work; instead, a new Device must be instantiated via new Device(token, options?).

As Connection is an overloaded and ambiguous term, the class has been renamed Call to better indicate what the object represents and be more consistent with Mobile SDKs and our REST APIs.

Device.setup() has been removed, and new Device(...) will not automatically begin connecting to signaling. There is no need to listen for Device.on('ready'). Instead, the signaling connection will automatically be acquired in one of two scenarios:

1. The application calls Device.connect(), creating an outbound Call. In this case, the state of the signaling connection will be represented in the Call.
2. The application calls Device.register(), which will register the client to listen for incoming calls at the identity specified in the AccessToken.

As long as outgoing calls are expected to be made, or incoming calls are expected to be received, the token supplied to Device should be fresh and not expired. This can be done by setting a timer in the application to call updateToken with the new token shortly before the prior token expires. This is important, because signaling connection is lazy loaded and will fail if the token is not valid at the time of creation.

Example:

1
const TTL = 600000; // Assuming our endpoint issues tokens for 600 seconds (10 minutes)
2
const REFRESH_TIMER = TTL - 30000; // We update our token 30 seconds before expiration;
3
const interval = setInterval(async () => {
4
  const newToken = await getNewTokenViaAjax();
5
  device.updateToken(newToken);
6
}, REFRESH_TIMER);

The Device states have changed. The states were: [Ready, Busy, Offline]. These have been changed to more accurately and clearly represent the states of the Device. There are two changes to Device state:

1. The states themselves have changed to [Registered, Registering, Unregistered, Destroyed]. This removes the idea of "Busy" from the state, as technically the Device can have an active Call whether it is registered or not, depending on the use case. The Device will always start as Unregistered. In this state, it can still make outbound Calls. Once Device.register() has been called, this state will change to Registering and finally Registered. If Device.unregister() is called the state will revert to Unregistered. If the signaling connection is lost, the state will transition to Registering or `Unregistered' depending on whether or not the connection can be re-established.

The destroyed state represents a Device that has been "destroyed" by calling Device.destroy. The device should be considered unusable at this point and a new one should be constructed for further use.

2. The busy state has been moved to a Boolean, Device.isBusy. This is a very basic shortcut for the logic return !!device.activeConnection.

The events emitted by the Device are represented by the Device.EventName enum and represent the new Device states:

1
export enum EventName {
2
  Destroyed = 'destroyed',
3
  Error = 'error',
4
  Incoming = 'incoming',
5
  Unregistered = 'unregistered',
6
  Registering = 'registering',
7
  Registered = 'registered',
8
}

Note that unregistered, registering, and registered have replaced offline and ready. Although frequently used to represent connected or disconnected, ready and offline actually were meant to represent registered and unregistered, which was quite ambiguous and a primary reason for the change.

When the device is destroyed using Device.destroy, a "destroyed" event will be emitted.

The construction signature and usage of Device has changed. These are the new API signatures:

1
/**
2
 * Create a new Device. This is synchronous and will not open a signaling socket immediately.
3
 */
4
new Device(token: string, options?: Device.Options): Device;
5

6
/**
7
 * Promise resolves when the Device has successfully registered.
8
 * Replaces Device.registerPresence()
9
 * Can reject if the Device is unusable, i.e. "destroyed".
10
 */
11
async Device.register(): Promise<void>;
12
/**
13
 * Promise resolves when the Device has successfully unregistered.
14
 * Replaces Device.unregisterPresence()
15
 * Can reject if the Device is unusable, i.e. "destroyed".
16
 */
17
async Device.unregister(): Promise<void>;
18
/**
19
 * Promise resolves when signaling is established and a Call has been created.
20
 * Can reject if the Device is unusable, i.e. "destroyed".
21
 */
22
async Device.connect(options?: Device.ConnectOptions): Promise<Call>;

The arguments for Device.connect() and Call.accept() have been standardized to the following options objects:

1
interface Call.AcceptOptions {
2
  /**
3
   * An RTCConfiguration to pass to the RTCPeerConnection constructor.
4
   */
5
  rtcConfiguration?: RTCConfiguration;
6

7
  /**
8
   * MediaStreamConstraints to pass to getUserMedia when making or accepting a Call.
9
   */
10
  rtcConstraints?: MediaStreamConstraints;
11
}

1
interface Device.ConnectOptions extends Call.AcceptOptions {
2
 /**
3
  * A flat object containing key\:value pairs to be sent to the TwiML app.
4
  */
5
  params?: Record<string, string>;
6
}

Note that these now take a MediaStreamConstraints rather than just the audio constraints. For example:

device.connect({ To: 'client:alice' }, { deviceId: 'default' });

might be re-written as:

1
device.connect({
2
  params: { To: 'client:alice' },
3
  rtcConstraints: { audio: { deviceId: 'default' } },
4
});

For backward compatibility, the new error format was attached to the old format under error.twilioError:

1
class oldError extends Error {
2
  //...
3
  code: number;
4
  message: string;
5
  twilioError: TwilioError;
6
}

The new Error format is:

1
class TwilioError extends Error {
2
  /**
3
   * A list of possible causes for the Error.
4
   */
5
  causes: string[];
6

7
  /**
8
   * The numerical code associated with this Error.
9
   */
10
  code: number;
11

12
  /**
13
   * A description of what the Error means.
14
   */
15
  description: string;
16

17
  /**
18
   * An explanation of when the Error may be observed.
19
   */
20
  explanation: string;
21

22
  /**
23
   * Any further information discovered and passed along at run-time.
24
   */
25
  message: string;
26

27
  /**
28
   * The name of this Error.
29
   */
30
  name: string;
31

32
  /**
33
   * The original Error received from the external system, if any.
34
   */
35
  originalError?: Error;
36

37
  /**
38
   * A list of potential solutions for the Error.
39
   */
40
  solutions: string[];
41
}

With the transition, the following error codes have changed:

• 31003 -> 53405 | When ICE connection fails
• 31201 -> 31402 | When getting user media fails
• 31208 -> 31401 | When user denies access to user media
• 31901 -> 53000 | When websocket times out in preflight

Previously, Device.setup() could only be used the set options once. Now, we've added Device.updateOptions(options: Device.Options) which will allow changing the Device options without instantiating a new Device. Note that the edge cannot be changed during an active Call.

Example usage:

1
const options = { edge: 'ashburn' };
2
const device = new Device(token, options);
3

4
// Later...
5

6
device.updateOptions({ allowIncomingWhileBusy: true });

The resulting (non-default) options would now be:

1
{
2
  allowIncomingWhileBusy: true,
3
  edge: 'ashburn',
4
}

This function will throw with an InvalidStateError if the Device has been destroyed beforehand.

The SDK now uses the loglevel module. This exposes several new features for the SDK, including the ability to intercept log messages with custom handlers and the ability to set logging levels after instantiating a Device. To get an instance of the loglevel Logger class used internally by the SDK:

1
import { Logger as TwilioClientLogger } from '@twilio/voice-client-sdk';
2
...
3
TwilioClientLogger.setLogLevel('DEBUG');

Please see the original loglevel project for more documentation on usage.

• Removed Connection.mediaStream. To access the MediaStreams, use Connection.getRemoteStream() and Connection.getLocalStream()
• Removed Connection.message in favor of the newer Connection.customParameters. Where .message was an Object, .customParameters is a Map.
• Removed the following private members from the public interface:
  • Connection.options
  • Connection.pstream
  • Connection.sendHangup
• Fixed Connection.on('cancel') logic so that we no longer emit cancel in response to Connection.ignore().

Some deprecated Device options have been removed. This includes:

• enableIceRestart
• enableRingingState
• fakeLocalDtmf

The above three removed options are now assumed true. The new Device.Options interface is now:

1
export interface Options {
2
  allowIncomingWhileBusy?: boolean;
3
  appName?: string;
4
  appVersion?: string;
5
  audioConstraints?: MediaTrackConstraints | boolean;
6
  closeProtection?: boolean | string;
7
  codecPreferences?: Connection.Codec[];
8
  disableAudioContextSounds?: boolean;
9
  dscp?: boolean;
10
  edge?: string[] | string;
11
  forceAggressiveIceNomination?: boolean;
12
  maxAverageBitrate?: number;
13
  rtcConfiguration?: RTCConfiguration;
14
  sounds?: Partial<Record<Device.SoundName, string>>;
15
}

The formula used to calculate the mean-opinion score (MOS) has been fixed for extreme network conditions. These fixes will not affect scores for nominal network conditions.