Smooth Out Network Chaos

Jitter buffer compensează variațiile în timpul de livrare a pachetelor, asigurând audio continuu și clar pentru voice AI.

What is Jitter?

Packet Arrival Without Jitter

Packets arrive at regular intervals (20ms each)

Packet Arrival With Jitter

Packets arrive at varying intervals, some out of order

Jitter = variation in packet arrival time. Measured in milliseconds. High jitter causes choppy, distorted audio.

How Jitter Buffer Works

Collect Incoming Packets

Store packets as they arrive from network

Reorder by Sequence Number

Put packets in correct order (RTP sequence)

Wait for Late Packets

Hold buffer to allow late arrivals to catch up

Release at Constant Rate

Output smooth, continuous audio stream

Buffer Types

Static (Fixed) Buffer

Fixed delay regardless of network conditions.

Buffer size:60ms (fixed)

Pros:Predictable, simple

Cons:Unnecessary delay on good networks

Adaptive Buffer (Recommended)

Adjusts size based on measured jitter.

Buffer range:20-200ms

Pros:Lower latency when possible

Cons:More complex, brief glitches during adaptation

Adaptive Buffer Algorithm

// Adaptive jitter buffer pseudocode
class AdaptiveJitterBuffer {
  minBuffer = 20;   // ms
  maxBuffer = 200;  // ms
  targetBuffer = 40; // initial

  onPacketReceive(packet) {
    const jitter = measureJitter(packet);

    // Expand buffer if jitter increases
    if (jitter > this.targetBuffer * 0.8) {
      this.targetBuffer = Math.min(
        this.maxBuffer,
        jitter * 1.5
      );
    }

    // Shrink buffer if network stabilizes
    if (jitter < this.targetBuffer * 0.5) {
      this.targetBuffer = Math.max(
        this.minBuffer,
        this.targetBuffer * 0.9
      );
    }

    this.buffer.add(packet);
  }
}

Buffer Metrics

35ms

Current Buffer

Adaptive target

12ms

Avg Jitter

Last 100 packets

0.02%

Underruns

Buffer empty events

0.1%

Discarded

Too late packets

Packet Loss Concealment

When packets are lost or arrive too late, PLC generates replacement audio:

Silence Insertion

Insert silence. Simple but noticeable.

Quality: Low

Packet Repetition

Repeat last packet. Works for short gaps.

Quality: Medium

Interpolation

AI predicts missing audio. Best quality.

Quality: High

Configuration Guidelines

Network Quality	Typical Jitter	Recommended Buffer	Mode
Excellent (LAN, Fiber)	<5ms	20-30ms	Adaptive (aggressive)
Good (Broadband)	5-20ms	40-60ms	Adaptive (normal)
Fair (Mobile 4G)	20-50ms	60-100ms	Adaptive (conservative)
Poor (Congested/3G)	>50ms	100-200ms	Fixed (high)

Voice AI Specific Considerations

STT Input Buffer

Smaller buffer preferred - STT can handle some jitter. Target: 20-40ms to minimize transcription delay.

TTS Output Buffer

Larger buffer acceptable - smooth playback prioritized. Target: 40-60ms for glitch-free AI speech.

Smooth Audio, Always

Jitter buffer pentru conversații fără întreruperi.

Vezi Demo →

Jitter Buffer