Tuning RƎLAY for low-latency VoIP
Optimizing RƎLAY’s buffer settings for low-latency VoIP callers is a balance between smooth speech and responsiveness. The buffer holds arriving packets briefly so network jitter does not become clipped words or late audio. A larger cushion absorbs irregular arrivals but adds delay. The target is the shortest buffer that remains stable on the actual route.
Delay is not created by RƎLAY alone. Capture, codec, packetisation, network routing, workstation output and station processing all contribute. Buffering mainly absorbs arrival variation; it cannot remove codec delay, congestion or packet loss. Tuning should therefore begin with measurements, not with the largest available setting.
Australian stations often connect contributors across varied paths. A caller in Sydney may join a program produced in Melbourne over enterprise services, while a regional broadcaster may rely on fixed wireless or a mobile backup. Public, commercial and community radio teams may all host remote talent, but the last mile can differ sharply between metropolitan, outer-suburb and regional locations.
Geography alone does not create internet latency; the routed path does. Interstate routes, shared uplinks and peak congestion can still make a stable profile fail. A useful setup accounts for those changes, separates delay in each direction and preserves margin for live talkback.
Build a stable baseline
The comparison separates three practical profiles for the same caller and processing chain.
| Buffer profile | Network behaviour | Expected result | Main trade-off |
|---|---|---|---|
| Short | Stable wired or well-managed route | Lowest possible delay | More exposed to late packets |
| Balanced | Ordinary VoIP with modest jitter | Clear speech with moderate delay | Requires periodic route checks |
| Extended | Variable WAN, Wi-Fi or cellular connection | Better tolerance of arrival peaks | Can feel sluggish |
Start with a controlled caller, codec, sample rate and processing chain, then keep them fixed while testing buffer changes. The RƎLAY virtual radio tools provide the production context, but the baseline should be a known-good call rather than a generic default. Record delay, jitter, packet loss and late-packet reports from available call or network diagnostics.
Treat these profiles as behaviours, not universal presets. Use the lowest matching profile, then increase it only when tests show recurring late packets. If a control is shown in samples, a 48 kHz engine contains 48 samples per millisecond, while a 96 kHz engine contains 96. Packet-based values require a different conversion, so the active RƎLAY version and measurement units should be checked before tuning.
Measure the real delay
Measure perceived delay from mouth to ear and from studio to caller rather than trusting application statistics alone. Calibrated monitoring can reveal added buffering, while network traces show whether variation occurs upstream. Keep loudspeaker acoustic delay out of the measurement loop because it can distort the result.
Change one parameter at a time. Move from the default in small increments, repeat a scripted passage with pauses and plosives, and note both intelligibility and delay. A slightly larger buffer is preferable if it prevents recurring late packets, but an oversized setting creates the sluggishness listeners notice first.
Test at different times and on the fallback route used for a mobile contribution. Weekday morning and evening tests can expose congestion affecting connections between Australian capital cities. Record the point at which words begin arriving late, not merely when audible artefacts appear.
Tune for changing routes
Maintain separate profiles for predictable studio links, ordinary hosted VoIP and less stable cellular or consumer broadband paths. Assigning the same value everywhere hides the cause of instability and makes a temporary network problem look like a permanent RƎLAY setting issue.
Where available, prioritise the VoIP traffic, reduce competing uploads and use the network’s supported quality-of-service controls. These steps can reduce delay and jitter but cannot guarantee a route. Packet loss, heavy forwarding or an overloaded uplink needs correction; a larger buffer can conceal it only for a limited time.
Do not equate codec packet duration with total jitter-buffer delay. A 20-millisecond packet interval may still require a much longer receive cushion on a variable path. Likewise, forward error correction can protect a weak link while adding send delay, so both directions must be judged.
Protect the live workflow
Protect live use with a dependable route and a documented fallback. Monitor caller audio, late packets, dropped frames and processing load before changing settings during an event. A local talkback or wired contribution can remain available if the VoIP path degrades, preserving the station’s on-air workflow rather than relying on software repair alone.
Physical studio boundaries still matter when VoIP feeds broadcast hardware. Cavendish audio hardware can support routing and interface choices that provide predictable conversion around the desk. Its role is complementary: buffering conditions the network stream, while physical routing determines whether that stream reaches the correct mix with stable timing.
Coordinate with the wider studio
Buffer changes belong within the wider automation context. Lawo’s explanation of VisTool automation integration shows why software state, physical control and existing station workflows must remain aligned. Version changes, sample-rate changes or automation recalls can invalidate assumptions that were valid during an earlier call.
Keep a show profile with the RƎLAY version, codec, sample rate, measured route and chosen buffer recorded beside it. Separate production and fallback profiles, then verify them after software updates. This makes support diagnosis faster and prevents a setting chosen for one Australian studio from being copied blindly to another.
Keep a known-good profile
The final setting should be treated as route-specific operating data. It should handle ordinary peak-period traffic, retain margin for normal jitter and avoid introducing enough delay to disrupt a live interview. Document exceptions when an event requires another profile, then restore the known-good value afterwards.
The value to remember is simple: the best buffer is neither the smallest number nor the largest cushion. It is the shortest receive buffer that keeps speech intact across the real call path in both directions.
