Neural Audio Codec Packet Loss Concealment via Language Model Prediction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing audio codec systems struggle with packet loss during real-time audio communication, leading to disruptions and distortions in the audio stream.

Innovation Solution

The integration of a nano Language Model (LM) Predictor within a neural audio codec system, which predicts missing codeword indices in the quantized domain to generate a seamless audio output.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional packet loss concealment methods are used in audio codec systems, then the system can handle packet loss, but the audio quality deteriorates with disruptions and distortions

Engineering Contradiction:

Improvepacket loss handling capabilityVSAvoidaudio quality

Core Design Contradiction:

ReliabilityVSManufacturing precision

Solution Approach 1:

A language model is introduced as an intermediary component between the audio decoder and the output. The language model predicts missing codeword indices based on the sequence of received packets, acting as a mediator that reconstructs lost information in a way that maintains audio quality while handling packet loss effectively

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system performs preliminary prediction of missing codeword indices using the language model before the audio decoding is completed. By predicting the lost indices in advance based on the language model and received packet sequences, the system prepares replacement data that maintains audio quality without causing disruptions

Inventive Principle:

Principle #10Preliminary action

2Manufacturing precision

If packet loss concealment is implemented to maintain audio quality, then audio disruptions are reduced, but the system complexity increases

Engineering Contradiction:

Improveaudio qualityVSAvoidconcealment system complexity

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The language model serves multiple functions: it models the language structure of codeword sequences, predicts missing indices, and adapts to different packet loss patterns. This multi-functionality reduces the need for separate specialized components, thereby managing system complexity while maintaining audio quality

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system changes the parameter space by working with codeword indices in the quantized domain rather than directly in the audio signal domain. This parameter transformation allows the language model to operate on discrete symbolic sequences, simplifying the prediction task while effectively concealing packet loss in the audio output

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20250131933A1Packet loss concealment in an audio decoder

Publication Date: 2025.04.24 CISCO TECHNOLOGY INC

AI summary

A method of performing packet loss concealment in a neural audio encoder/decoder (codec) system. The method includes receiving an indication of a lost audio packet at a receive side of a neural network audio codec system that includes an audio encoder and an audio decoder, wherein the lost audio packet comprises an index of a codeword that is representative of a portion of speech audio presented to the audio encoder, predicting the index of the codeword in the lost packet to obtain a predicted index, deriving a predicted embedding vector from the predicted index, and decoding, by the audio decoder, the embedding vector to generate an audio output.