Digital audio in a call is sent as a sequence of small packets, each carrying a few tens of milliseconds of sound. For the listener to hear a smooth voice, those packets must be played at the same steady pace at which they were captured. The network does not preserve that rhythm. Packets take different routes or wait in different queues, so they arrive unevenly, some early and some late, and occasionally out of order. This variation in arrival time is jitter. A listener does not perceive it directly. They hear its effects as clicks, gaps, robotic distortion or words that cut out. Because speech is highly sensitive to interruption, handling jitter well is one of the central problems of real-time audio systems.
What is the impact of jitter on audio quality?
If the receiver plays each packet the moment it arrives, uneven arrival becomes uneven sound. A packet that arrives too late to be played misses its slot, which has the same effect as a lost packet, leaving a hole in the audio. Even if the packet arrives eventually, it is useless by then. Small amounts of jitter produce subtle roughness and larger amounts produce audible dropouts. Jitter also tends to appear together with congestion, so it is often accompanied by genuine loss. The effect varies with the codec and the content, since steady vowels are easier to conceal than sudden consonants, and music exposes problems that speech may hide.
How do jitter buffers work?
The standard remedy is a jitter buffer at the receiver. Instead of playing packets immediately, it holds them briefly and releases them at a regular pace, using sequence numbers and timestamps carried in the Real-time Transport Protocol to restore the original order and timing. The cost is delay, since every packet waits at least as long as the buffer depth. A fixed buffer is simple, and it must be sized for the worst expected jitter, which adds unnecessary delay in good conditions. Adaptive buffers monitor the recent variation and grow or shrink accordingly, stretching or compressing the audio slightly when they adjust. The aim is the smallest delay that still avoids most late packets, which matters in conversation, where long delay makes people talk over each other.
Concealment and other mitigations
Even a well-tuned buffer cannot recover a packet that never arrives. Packet loss concealment algorithms generate plausible audio to cover the gap, for example by extending the last waveform, and modern codecs such as Opus have concealment built in. Forward error correction sends redundant information so that some losses can be repaired without a retransmission, and a few codecs carry a low-quality copy of the previous frame within the next packet. Network measures help too. Marking audio packets with a priority class lets networks that support quality of service treat them first, and reducing competing traffic or using wired connections cuts jitter at its source. Monitoring jitter and loss statistics reported by the protocol shows where problems originate.
Why it matters
Voice and audio communication depends on small time scales, and a network designed for bulk data does not deliver them naturally. The combination of an adaptive buffer, concealment and sensible network configuration is what lets calls sound natural across imperfect connections. Understanding the trade between delay and smoothness is the key to diagnosing and improving real-time audio.
Every jitter buffer is a bet that a little delay now is better than a gap later.