CELP Speech Decoding Lost Frame Concealment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech codecs, particularly CELP speech codecs, face challenges in maintaining high-quality decoding when frame loss occurs, especially in voiced portions where adaptive codebook usage leads to distortion, and existing concealment methods struggle with predictive encoding methods that rely on past encoded information.
Innovation Solution
A speech decoding apparatus that generates an average waveform pattern of excitation signals across multiple frames and uses this pattern to create a concealed frame for lost frames, improving concealment performance and decoded speech quality by employing a configuration with a decoding section, generation section, and concealment section.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If CELP speech codec uses adaptive codebook with predictive encoding, then encoding efficiency is improved, but decoded speech quality deteriorates when frame loss occurs
Solution Approach 1:
The patent prepares concealment excitation signals in advance by storing unquantized excitation signals from previous frames. When frame loss occurs, these pre-prepared signals are immediately used for concealment, avoiding the need to regenerate them during loss events and maintaining speech quality despite the predictive encoding dependency.
Solution Approach 2:
The patent introduces an intermediary concealment mechanism that uses unquantized excitation signals as a bridge between the lost frame and the decoded output. This intermediary signal allows the system to bypass the broken predictive encoding chain and directly synthesize plausible speech content without being constrained by the failed adaptive codebook references.
2Device complexity
If conventional concealment uses preceding frame information, then simple implementation is achieved, but concealment effectiveness deteriorates in speech onset portions
Solution Approach 1:
The patent applies different concealment strategies based on the local characteristics of the speech signal. For onset portions where correlation is low, it uses unquantized excitation signals that capture the transient characteristics better. For steady-state portions, conventional methods can be used, optimizing concealment quality for each local region of the speech signal.
Solution Approach 2:
The patent changes the parameter representation from quantized codebook indices to unquantized excitation signals. This parameter change preserves more information about the actual excitation waveform, enabling better concealment of onset portions where the excitation characteristics are critical and cannot be adequately represented by coarse codebook indexing.
3Reliability
If loss concealment transmits encoded information for preceding/succeeding frames, then concealed signal quality is improved, but transmission bandwidth requirement increases
Solution Approach 1:
The patent extracts only the essential concealment information (unquantized excitation signals from previous frames) and transmits it separately from the main encoded speech data. This extraction approach provides the necessary concealment capability without transmitting the full preceding frame encoded information, reducing the bandwidth overhead while maintaining concealed signal quality.
Data Source
AI summary
A sound decoding device is capable of improving the lost frame compensation performance and improving quality of the decoded sound. A rise frame sound source compensation unit generates a compensation sound source signal when the current frame is a lost frame and a rise frame. An average sound source pattern update unit updates the average sound source pattern held in an average sound source pattern holding unit over a plurality of frames. When a frame is lost, an LPC synthesis unit performs LPC synthesis on a decoded sound source signal by using the compensation sound source signal inputted via a switching unit and a decoded LPC parameter from an LPC decoding unit and outputs the compensation decoded sound signal.