Twilio Voice SDK: Mobile Voice Calls Integration
The user taps the call button — but the call doesn't go through. Or an incoming call arrives with a 10-second delay. Most often the problem lies in incorrect Twilio Voice integration: Access Token expired, PushKit not configured, CallKit invoked in the wrong order. Let's break down how to avoid these errors.
Twilio Programmable Voice SDK is an industrial platform that eliminates the hassle of TURN servers and media servers. Twilio handles media transport (Opus/DTLS/SRTP) and global routing through 90+ data centers. The developer is left with: signaling through Twilio, managing the audio session on the mobile platform, and integration with system call interfaces (CallKit on iOS, ConnectionService on Android). We have 5+ successful Twilio Voice projects under our belt, guaranteeing stable operation in production.
Server-Side Setup: Access Token
Twilio Voice SDK authenticates via short-lived Access Tokens generated by your backend using the Twilio Helper Library. The token contains a VoiceGrant — permissions for incoming and outgoing calls. Typical configuration:
from twilio.jwt.access_token import AccessToken
from twilio.jwt.access_token.grants import VoiceGrant
token = AccessToken(
account_sid=ACCOUNT_SID,
signing_key_sid=API_KEY_SID,
private_key=API_KEY_SECRET,
identity=user_id,
ttl=3600
)
token.add_grant(VoiceGrant(
outgoing_application_sid=TWIML_APP_SID,
incoming_allow=True
))
return token.to_jwt()
The mobile client fetches the token on launch and renews it before expiry. The Twilio SDK notifies via the delegate accessTokenInvalidOrExpired — at this point you need to request a new token and call updateAccessToken. We recommend a TTL of 3600 seconds — a balance between security and refresh frequency. Twilio official documentation confirms this approach.
Steps to generate an Access Token on the server:
- Create an API Key in Twilio Console.
- Configure a TwiML Application with a webhook URL.
- Write an endpoint in Node.js/Python that returns a JWT with VoiceGrant.
Why PushKit is Mandatory for Incoming Calls on iOS
PushKit is the only reliable way to deliver an incoming call in any application state. The APNs VoIP channel guarantees priority delivery. Without PushKit, the app won't receive the signal in the background.
func pushRegistry(_ registry: PKPushRegistry,
didUpdate credentials: PKPushCredentials,
for type: PKPushType) {
TwilioVoice.register(accessToken: token,
deviceToken: credentials.token) { error in }
}
func pushRegistry(_ registry: PKPushRegistry,
didReceiveIncomingPushWith payload: PKPushPayload,
for type: PKPushType,
completion: @escaping () -> Void) {
TwilioVoice.handleNotification(payload.dictionaryPayload,
delegate: self,
delegateQueue: nil)
// MUST call CallKit reportNewIncomingCall before completion
}
Violating the rule of calling CallKit before completion in the PushKit delegate results in forced app termination by iOS. This is not a warning, it's a crash. Integration with CallKit via TVODefaultAudioDevice — Twilio provides a ready-made AVAudioSession manager that correctly interacts with CallKit.
How to Connect ConnectionService on Android
Dependency: com.twilio:voice-android:6.x.x. The SDK works on top of WebRTC but provides a high-level API.
// Initialization
Voice.initialize(context, LogLevel.DEBUG)
// Outgoing call
val connectOptions = ConnectOptions.Builder(accessToken)
.params(mapOf("To" to phoneNumber))
.build()
val call = Voice.connect(context, connectOptions, object : Call.Listener {
override fun onConnected(call: Call) { /* call established */ }
override fun onDisconnected(call: Call, error: CallException?) { /* ended */ }
override fun onConnectFailure(call: Call, error: CallException) { /* error */ }
})
Incoming calls come via FCM. The Twilio SDK handles FCM payload through Voice.handleMessage():
override fun onMessageReceived(message: RemoteMessage) {
if (Voice.handleMessage(context, message.data, object : MessageListener {
override fun onCallInvite(callInvite: CallInvite) {
// show incoming call notification
showIncomingCallNotification(callInvite)
}
override fun onCancelledCallInvite(cancelledInvite: CancelledCallInvite, ...) {
// call cancelled before answer
}
})) { /* this is a Twilio push */ }
}
Answering a call: callInvite.accept(context, callListener). To work correctly with ConnectionService, you must specify the correct android:name in the manifest — otherwise the system won't be able to activate the service.
TwiML and Server-Side Routing
Note: when a mobile client calls via Twilio, the request hits your TwiML Application webhook. The backend responds with TwiML — XML instructions for Twilio:
<?xml version="1.0" encoding="UTF-8"?>
<Response>
<Dial callerId="+1234567890">
<Client>recipient_user_id</Client>
</Dial>
</Response>
For calls to regular phone numbers, use <Number> instead of <Client>. Twilio acts as an intermediary, your server controls routing logic.
Call Recording and Analytics
Recording via TwiML <Record> or programmatically via REST API — available without SDK changes. Twilio stores recordings on its servers, providing a download URL. Call analytics (duration, quality, statuses) — via Twilio Console or REST API. We configure automatic notification of recordings to your CRM.
Twilio Voice vs. Custom WebRTC
| Criterion | Twilio Programmable Voice | Custom WebRTC + TURN |
|---|---|---|
| Time to production | 1–3 weeks | 3–6 months |
| Infrastructure | Zero (ready) | TURN server, media server, WebSocket |
| Additional latency | < 50 ms (closest PoP) | 0 ms (P2P) |
| Call recording | Built-in | Requires development |
| CallKit/ConnectionService support | Built-in | Requires implementation |
| Cost per minute | from $0.014 | $0.001–0.005 (only TURN) |
When to Choose Custom WebRTC?
Twilio Voice adds latency through relay — all media streams go through Twilio data centers, not P2P. For most tasks this is negligible (< 50 ms additional latency at the closest PoP), but in regions without a nearby Twilio data center (Central Asia, parts of Africa) latency may be noticeable. Cost: at high volumes (10,000+ minutes/day) custom WebRTC + TURN is cheaper, but more expensive in development and maintenance. If you need P2P communication without intermediaries — contact us, we'll help you choose the architecture.
Integration Process
| Step | Actions | Duration |
|---|---|---|
| 1. Twilio account setup | TwiML App, API Keys, Push Credentials (FCM/APNs) | 1 day |
| 2. Backend implementation | Access Token generation, TwiML webhook (Node.js, Python, Go) | 2–5 days |
| 3. SDK integration | Android (Kotlin) + ConnectionService, iOS (Swift) + CallKit | 3–7 days |
| 4. Testing | Real devices, bug fixing | 2–3 days |
Full integration timeline — 1–3 weeks. Infrastructure savings — up to $50,000 in the first year.
What's Included
- Backend development and setup: Access Token generation, TwiML webhook (Node.js or Python).
- Twilio Voice SDK integration on iOS (Swift, CallKit) and Android (Kotlin, ConnectionService).
- PushKit (iOS) and FCM (Android) configuration for incoming calls.
- Testing on real devices, bug fixing.
- Integration documentation and operations manual.
- Team training (up to 2 hours online).
Contact us for a consultation on architecture and pricing. Order Twilio Voice integration from us – get stable calls in 2 weeks.







