Run your voice services
Play announcements, collect DTMF and connect conferences. Your IMS application server controls the media through SIP, NETANN, MSML or VoiceXML.
Keep service logic in your application server.
A subscriber calls your service. They hear an announcement, choose an IVR option or speak to an assistant. Behind each step, your network needs the right media, in the codec negotiated for that call.
Skylark Media Server brings announcements, conferencing, encrypted recording and speech-engine integration into one platform. In IMS, it provides the Media Resource Function while your application server controls the service.
Start with the services your subscribers use today. Add recording or speech processing where the call flow needs them, on the same media platform.
Play announcements, collect DTMF and connect conferences. Your IMS application server controls the media through SIP, NETANN, MSML or VoiceXML.
Keep service logic in your application server.
Capture each call side on its own channel. Skylark encrypts audio in memory and sends recordings to encrypted storage, without writing an unencrypted copy to disk.
Preserve separate channels for downstream review.
Fork media to recognition while the call continues. Return synthesised speech in the negotiated call codec, using native or external speech engines.
Add voice AI through the network media path.
Skylark provides the Media Resource Function (MRF), including MRFC and MRFP. SIP carries media-control requests on Mr and Mr′; RTP carries the call media on Mb.
A subscriber calls your care service. Follow the prompt, their keypad choice and the speech path behind the conversation.
The call stages play automatically while the phone is visible. Drag or use arrow keys to rotate the phone. Enter or Space continues the call flow. On the keypad, press 1 to select account support. R, Home or a double click resets the view.
Connecting the call. Your application server requests media over SIP. Skylark establishes the RTP path.
A call moves between network routes, service prompts and the people or applications behind them. Skylark handles the media at each step, while your teams control how the service runs.
A call arrives in EVS or AMR-WB, but the service needs another format. Skylark transcodes between the negotiated call codec and that service format, keeping media processing aligned with the call.
Existing services depend on specific media controls and scripts. Skylark supports VoiceXML 2.0/2.1; map each application’s dependencies and test the operations and responses it needs before cutover.
Some services need a conference or a fax session beyond an IVR menu. Skylark provides conferencing and T.38 fax; test those paths with the actual endpoints and routes your service will use.
Video brings its own codec requirements into a call. Skylark supports H.264 and H.265, plus VP8/VP9 for WebRTC, so those paths can be included alongside voice in your service acceptance tests.
A caller can speak instead of choosing from a keypad menu. Skylark handles the media for recognition and synthesis; AI reasoning and call treatment remain with the service controlling the conversation.
When a media problem reaches the NOC, the team needs operational evidence. Skylark exposes SNMPv2/3 and Prometheus/Grafana metrics, bringing media visibility into your existing OAM tools.
A call-screening service can ask an unfamiliar caller for the context of their call. That exchange needs media in both directions: the caller's response goes to recognition, and the next spoken prompt returns to the same voice session.
Skylark provides that audio path. The application server and AI layer determine the call treatment, so the media exchange remains separate from the decision about how the call should proceed.
Skylark combines IMS media with a native speech option and Kubernetes scaling. Compare the capabilities and the integration work around them.
| Decision area | Dialogic PowerMedia XMS | Radisys Engage Media Server | Skylark Media Server |
|---|---|---|---|
| Speech-engine ownership | Third-party ASR/TTSExternal speech provider to integrate | Built-in keyword detectionFull ASR/TTS uses third-party engines | Native or external ASR/TTSRun speech within Skylark, or use your engine |
| Live speech integration | MRCP / cloud providersGoogle/Amazon STT: file-based | MRCP v1 / v2Third-party speech-engine endpoint | Live PCM over WebSocketSame Skylark interface across engines |
| Scale-out operations | VMs / containers / cloudAdditional MRB for pooled scale and HA | Kubernetes CNF / VM VNFContainer or VM deployment | Kubernetes CNF / VNFReplicas follow call and session load |
| Recording protection | Stereo WAV captureEncrypted WebM / MKV formats | Encrypted recording / playback | Per-side; encrypted in memoryNo plaintext copy written to disk |
| VoLTE audio | EVS / AMR-WB; transcoding | EVS / AMR-WB; transcoding | EVS / AMR-WB; transcoding |
| Application control | MSML / NETANN / VoiceXML / REST | MSML / NETANN / VoiceXML / H.248 | MSML / NETANN / VoiceXMLReuse your IMS application control model |
Deploy Skylark as a Kubernetes CNF or a VNF. For the CNF, scale replicas to call and session load, and drain sessions during planned updates. Native Rust code runs without a garbage collector.
Start with one service on the existing application server. Validate its media controls and codec paths, then move further services with a defined rollback route.
For an IMS Data Channel integration, Skylark is designed to take the Media Function role in the NG-RTC architecture, linking Data Channel media with call audio for services such as live captions.
An in-call screen needs a compatible handset and IMS integration. Your voice AI service can run over VoLTE or VoNR without Data Channel support in the handset.
Move one service at a time. Agree the call flow, codec combinations and media controls, then validate those flows on Skylark before moving further services. Keep a defined rollback route to the existing media server. Include application-server routing and treatment of active sessions in the cutover plan.
Check the MSML packages used by each application rather than treating protocol support as full application compatibility. Skylark supports VoiceXML 2.0 and 2.1, alongside SIP and NETANN control. Test your actual announcements, DTMF collection and conference operations with the application server, and confirm the required MSML package coverage before cutover.
Size against your traffic profile: simultaneous sessions, codec pairs, conferencing, recording and the share of calls using speech engines. Validate that combination under representative load. Skylark scales replicas within configured limits, with a SIP distribution layer spreading calls across pods. Agree capacity and headroom for the service mix you intend to run.
Your integration has two boundaries: Skylark exchanges audio with the selected speech engine, while call control determines which sessions use AI and how the service continues. For acceptance, test recognition and synthesis in one end-to-end call, including the codec negotiated on the network route and the PCM/WebSocket handoff.
Agree how the two call channels are correlated with the call record, transferred to encrypted storage and made available to authorised downstream consumers. Validate replay and access through that storage integration as part of acceptance. This is separate from Skylark’s capture and in-memory encryption of the audio.
During a planned update, a drained pod closes after its last session ends or the configured time limit is reached. If a pod fails, Kubernetes restarts it and its sessions are released. Set the drain window around your call durations and test application-server treatment of interrupted sessions separately from rolling updates.
Define responsibility for SIP call distribution, replica limits, monitoring and the recording or speech integrations around the media server. Kubernetes handles CNF pod replacement, but service recovery also depends on how the application server treats released sessions. For a VNF, agree the corresponding capacity and recovery procedures.
Skylark supports EVS, AMR-WB, AMR-NB, G.711, G.722, G.729 and Opus. Test the codec pairs negotiated on your actual routes and transcoding to each service format. Cover video and T.38 fax separately. For voice AI, include both call audio to recognition and synthesised speech back to the caller.
Yes. Voice AI can run over VoLTE or VoNR without Data Channel support in the handset. An in-call screen needs a compatible terminal and the Data Channel integration. Plan that separately, including Skylark's Media Function role, DC2 media-resource control and the connections to the Data Channel signalling and application functions.
Bring your call flow, codecs and application-server interfaces. Together we can map the media path, agree recording and speech-engine integration, and plan a service-by-service migration.