What Is Latency in Audio over IP?

Latency in Audio over Internet Protocol (AoIP) systems is the total time it takes for an audio signal to travel from the source (microphone, mixer output, playback device) to its destination (amplifier, recorder, loudspeaker) through an IP network. Unlike analog or traditional digital audio (e.g., AES/EBU), AoIP introduces additional delays due to packetization, buffering, and network traversal. Even small amounts of latency—measured in milliseconds—can create perceptible problems in live sound reinforcement, broadcast intercom systems, or multi-track recording sessions where multiple audio streams must remain in tight alignment.

In real-world applications, latency manifests in several ways: a distant echo in a monitor wedge, lip-sync errors in video production, or a slight flamming effect when multiple musicians monitor their own delayed returns. Professional AoIP systems aim for round-trip latency below 1 millisecond for in-ear monitoring and under 5 milliseconds for general live audio distribution. Achieving these targets requires understanding the distinct layers where delay accumulates.

Types of Latency in AoIP

Algorithmic Latency

This is the time required for digital signal processing tasks such as sample rate conversion, filtering, or audio codec encoding/decoding. In uncompressed AoIP (e.g., AES67, Dante, RAVENNA), algorithmic latency is minimal, often just a few sample periods. However, if compression is used—such as Opus or MPEG AAC over IP—codec-induced delays can become a significant contributor, adding 5-30 milliseconds depending on the algorithm and frame size.

Buffering Latency

Every AoIP device uses buffers to collect audio samples before forming packets and to reorder incoming packets before playback. Buffers exist at both the transmitter and receiver. Transmit buffers wait for enough samples to fill a packet payload. Receive buffers (de-jitter buffers) store packets to smooth out network jitter. Larger buffers increase robustness against packet loss and timing variation but directly increase latency. A typical Dante receiver buffer of 1 millisecond contributes 1 ms one-way, while some devices allow buffer settings from 0.25 ms to 10 ms or more.

Network Propagation Latency

This includes the physical travel time of electrical or optical signals through cables and switches. In a local area network (LAN) with distances under 100 meters, propagation delay is negligible (under 1 microsecond per meter). However, in wide-area deployments with fiber optic links and multiple router hops, propagation latency can grow to several milliseconds. For most studio and live sound environments, this is the smallest factor.

Processing and Serialization Latency

Each switch, router, or media converter introduces processing delay—the time to read the packet header, make forwarding decisions, and (in managed switches) apply Quality of Service (QoS) rules. Store-and-forward switches also add serialization delay proportional to packet size: a 1,500-byte packet at 1 Gbps takes about 12 microseconds to serialize. While individually small, the cumulative effect across many switches can become measurable.

Key Factors That Influence AoIP Latency

Network Hardware and Infrastructure

The choice of Ethernet switches, routers, and cabling directly affects latency. Consumer-grade switches often use shared backplanes and store-and-forward processing that introduces more delay than professional-grade switches with cut-through forwarding. Managed switches with latency as low as 3-5 microseconds per port are available for audio-critical networks. Additionally, the cable type matters: Cat6 or Cat6a twisted-pair cabling supports 1 Gbps speeds and lower latency than Cat5e in high-bandwidth scenarios. Fiber optic connections eliminate electromagnetic interference but add a small propagation latency per kilometer.

Recommendation: Use switches specifically designed for AV-over-IP, such as those certified for Dante or AES67. These switches implement features like ingress rate limiting, priority queuing, and rapid spanning tree to keep latency deterministic. Audinate provides detailed switch configuration guides for optimal Dante performance.

Packet Processing and Buffering Strategies

Every AoIP node (encoder, decoder, microphone, amplifier) processes audio in blocks of samples. The block size determines how often packets are transmitted. A device using 1-millisecond packet intervals (48 samples at 48 kHz) will have lower algorithmic latency than one using 4-millisecond intervals (192 samples), but the smaller packet size increases network overhead and may stress switch processing. Buffer sizes on the receiving end must be large enough to accommodate network jitter without underruns. The trade-off between low latency and robustness is a fundamental design decision.

Best practice: For closed networks with minimal traffic, set receive buffers to 1 millisecond or less. For shared or congested networks, 2-5 milliseconds may be necessary. Some protocols like AES67 define mandatory latency modes (e.g., 1 ms, 2 ms, 5 ms) to ensure interoperability across brands.

Network Traffic and Bandwidth

AoIP systems are sensitive to congestion. When other traffic (file transfers, video streams, browsing) competes for bandwidth, packets can be queued in switch buffers, increasing latency unpredictably. Even with high total bandwidth, bursts of data can cause micro-delays. The solution is network segmentation: dedicate a VLAN for audio traffic, or use a physically separate network. QoS marking (DSCP values) can prioritize AoIP packets, but only if every device on the path honors the markings. RAVENNA and Dante both support QoS tagging, but many legacy switches ignore it.

Practical tip: Never run AoIP on a network that also handles large data transfers during critical sessions. If sharing is unavoidable, implement strict traffic shaping and limit non-AoIP bandwidth to 30% of the link capacity.

Clock Synchronization

All AoIP systems rely on a common clock reference to keep sample rates aligned. Precision Time Protocol (IEEE 1588) is used by Dante, AES67, and RAVENNA to distribute timing. Clock drift or poor clock recovery forces receivers to insert or drop samples (sample slipping), which introduces audible clicks or requires resampling buffers that add latency. High-quality network interface cards (NICs) and dedicated clock generators (e.g., a master clock feeding PTP) reduce timing error.

Key point: A properly synchronized AoIP network can achieve sub-microsecond timing accuracy. However, if the master clock is unstable or the network has excessive jitter, PTP will degrade and receivers may increase their buffer sizes automatically to compensate, raising latency. Sound On Sound’s guide on clocking explains how to set up a stable PTP domain.

Protocol Overhead and Media Transport

The choice of AoIP transport protocol influences latency in subtle ways. Dante uses a proprietary layer over IP, but its latency is configurable in fixed steps (0.25, 0.5, 1, 2, 5 ms). AES67, being an open standard, defines interoperable modes but may require additional conversion layers that add microseconds. RAVENNA offers extremely low latency (down to 0.125 ms) due to its efficient use of multicast and minimal buffering. The protocol also determines how packet loss is handled—some protocols use redundancy (e.g., redundant streams) while others rely on retransmission, which can add jitter and delay.

Note: For critical low-latency applications like in-ear monitoring, consider a protocol that supports redundant networks (e.g., Dante primary/secondary) to avoid retransmission delays from lost packets.

Jitter and Its Impact on Effective Latency

Jitter—the variation in packet arrival times—is often the hidden cause of higher latency. Even low average latency becomes unusable if jitter spikes force the receiver to lengthen its jitter buffer. Jitter sources include low-quality switch clocks, full-duplex mismatch, long cable runs without proper termination, and electromagnetic interference. Measuring jitter with tools like Wireshark or specialized network analyzers helps identify whether buffer sizes need to be increased or if the network requires remediation.

Rule of thumb: For every 1 microsecond of jitter, you may need 5-10 microseconds of buffer headroom to avoid dropouts. Reducing jitter at the source—by using better switches, proper cabling, and network load management—is more effective than simply increasing buffers.

Strategies to Minimize Latency

Select Low-Latency Network Hardware

Invest in managed switches with cut-through forwarding, low port-to-port latency (under 10 µs), and support for AVB or IEEE 802.1Qat stream reservation. Switches like the Netgear M4250 series or Cisco Catalyst 9300 are popular in professional audio installations. Ensure all switches in the path have the same capabilities to avoid bottlenecks. Avoid wireless bridges or consumer routers; they introduce unpredictable latency spikes.

Implement Quality of Service and Network Segmentation

Configure QoS to give AoIP traffic highest priority. Use DSCP values of 44 (CS5) or 46 (EF) for audio packets. Set strict priority queuing on all switches, and enable ingress policing to shape bursty traffic. Separating audio onto its own VLAN or dedicated physical network eliminates interference. If a single cable must carry both control (e.g., OSC, HTTP) and audio, separate VLANs with proper 802.1Q trunking are mandatory.

External resource: AtlasIED’s QoS guide for AV-over-IP provides configuration examples for major switch brands.

Optimize Buffer Sizes on Every Device

Manufacturers often default to larger buffers for reliability. For a stable, dedicated network, reduce transmit and receive buffers to the smallest value that avoids dropouts. Test with the lowest buffer setting and gradually increase until dropouts stop. Keep track of buffer settings for each device type—some digital mixers allow per-channel buffer adjustments. Use the protocol’s latency mode (e.g., Dante’s “Low Latency” mode) when available.

Choose the Right Protocol for the Application

For broadcast production requiring sub-0.5 ms round-trip, RAVENNA is typically preferred. For general live sound with up to 1 ms, Dante is ubiquitous and well-supported. For interoperability across multiple manufacturers, AES67 is essential but may not offer the lowest latency. Evaluate the maximum number of channels needed and whether the protocol supports redundant streams without extra latency.

Maintain a Clean Physical Layer

Replace old or untwisted cables. Use solid-core Cat6a for fixed installations—it has better signal integrity than stranded patch cables. Terminate all connections with TIA/EIA-568-B standards to avoid crosstalk. Keep cable runs under 100 meters per segment. Use shielded cable in environments with high RFI or EMI. Grounding and bonding of network devices also reduce jitter from electrical noise.

Monitor and Test Your Network

Regularly check latency using tools like iperf with UDP tests, or specialized software like Dante Controller’s latency monitor. Measure jitter with ping -f or PRTG. Document baseline latency values. When adding new devices, verify they don’t introduce extra delay. Audinate’s network basics include a latency validation procedure.

Use Professional Services for Commissioning

If your facility demands deterministic low latency across hundreds of channels, consider hiring a network integrator certified in AoIP systems (e.g., Dante Certified or AES67 certified). They can perform site surveys, switch configuration, and end-to-end latency measurements that save hours of troubleshooting.

Putting It All Together: A Practical Latency Budget

A typical low-latency Dante system might have the following breakdown:

  • Source to network TX: 0.5 ms (buffer + packetization)
  • Switch transit (3 switches): 0.03 ms (10 µs each)
  • Network propagation: 0.001 ms (10 meters cabling)
  • Receiver de-jitter buffer: 1.0 ms
  • DAC processing: 0.2 ms

Total one-way: approximately 1.731 ms. Adding the return path doubles that for round-trip latency to about 3.5 ms – well within the acceptable range for most live mixing. With careful selection of hardware and buffer tuning, round-trip latency can be pushed below 1 ms using RAVENNA or Dante in multicast mode with 0.25 ms buffers.

Conclusion

Minimizing latency in AoIP systems is not a single adjustment but a discipline applied across hardware, network configuration, protocol selection, and ongoing monitoring. By understanding where delays originate—buffering, network topology, clock jitter, and QoS handling—audio professionals can make informed decisions that preserve the real-time feel of analog audio while leveraging the flexibility of IP networks. Start with a clean network infrastructure, prioritize audio traffic, tune buffer settings for your environment, and validate with regular measurements. Following these best practices ensures that your AoIP system delivers the low latency and rock-solid synchronization that modern production demands.