Adaptive Speech Coding Scheme for Low Bit Rate Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech coding techniques at low bit rates often produce synthesized speech that sounds artificial, noisy, and overly harmonic, struggling to maintain toll-quality decoding with limited bit rates, especially for voiced frames.
Innovation Solution
A method of encoding speech signal frames by calculating peak and average energy, selecting between noise-excited and nondifferential pitch prototype coding schemes, and encoding frames with representations of pitch pulse shape, position, and estimated pitch period to optimize bit allocation and coding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If conventional low-bit-rate speech coding techniques are used, then the bit rate is reduced, but the synthesized speech quality deteriorates (sounds artificial, noisy, and overly harmonic)
Solution Approach 1:
The patent applies dynamics by making the coding scheme adaptable and variable rather than fixed. The encoder dynamically selects between noise-excited coding and nondifferential pitch prototype coding based on frame characteristics (voiced/unvoiced classification). This dynamic adaptation allows the system to optimize speech quality for different types of frames while maintaining low overall bit rate, resolving the contradiction between reduced bit rate and maintained speech quality.
Solution Approach 2:
The patent applies local quality by using different coding schemes for different portions of the speech signal. Voiced frames (which require higher quality) are encoded using nondifferential pitch prototype coding that preserves pitch pulse shape and position information, while unvoiced frames are encoded using noise-excited coding. This localized differentiation ensures high quality where needed while maintaining low overall bit rate.
2Manufacturing precision
If more bits are allocated to encode pitch and spectral information, then the speech quality improves, but the bit rate increases
Solution Approach 1:
The patent applies partial action by encoding only the most critical parameters for speech quality rather than attempting to encode the complete speech signal. For voiced frames, it encodes pitch pulse shape, position, and estimated pitch period - the essential elements for natural speech reproduction - while using fewer bits than would be required for full-bandwidth encoding. This selective encoding achieves good speech quality at low bit rate.
Solution Approach 2:
The patent applies parameter changes by transforming the encoding approach based on frame type. For voiced frames, it uses nondifferential pitch prototype coding that estimates pitch period and encodes pitch pulse characteristics. For unvoiced frames, it switches to noise-excited coding with different parameter sets. This parameter adaptation allows efficient bit allocation that maintains speech quality while controlling bit rate.
3Device complexity
If a single coding scheme is used for all frames, then the device complexity is reduced, but the speech quality for different frame types deteriorates
Solution Approach 1:
The patent applies dynamics by implementing a multi-mode coding structure where the encoder dynamically selects between noise-excited coding and nondifferential pitch prototype coding based on voice activity detection. This dynamic scheme classification allows the system to adapt to different frame types (voiced/unvoiced) and maintain high speech quality for each type while keeping the overall device complexity manageable through systematic mode selection and switching.
Data Source
Figure 1
Figure 2A~2B
Figure 3A~3B
AI summary
Systems, methods, and apparatus for low-bit-rate coding of transitional speech frames are disclosed.