Half-Rate MBE Vocoder for Low-Bitrate Voiced Speech Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech coders face challenges in encoding speech at low to medium bit rates, particularly in accurately representing voiced sounds and adapting to changes in speech signals, leading to reduced quality and intelligibility due to insufficient periodicity and sensitivity to noise.
Innovation Solution
The implementation of a Multi-Band Excitation (MBE) vocoder that uses a harmonic representation for voiced speech and a flexible voicing model to better accommodate mixed voicing sounds, allowing for improved quality and robustness to noise, with the half-rate MBE vocoder operating at 3600 bps achieving comparable performance to the standard 7200 bps APCO Project 25 vocoder.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If linear prediction coding is used to encode speech at low to medium bit rates, then time resolution is improved which helps in coding unvoiced sounds, but voiced sounds become rough or hoarse due to insufficient periodicity
Solution Approach 1:
The speech signal is divided into multiple frequency bands using a filter bank, with each band processed independently by a separate excitation source. This segmentation allows different excitation strategies to be applied to different frequency regions, improving both time resolution for unvoiced sounds and periodicity representation for voiced sounds in their respective bands.
Solution Approach 2:
The vocoder dynamically selects between different excitation types (noise-based for unvoiced, periodic for voiced) on a band-by-band basis. This dynamic adaptation allows the system to optimize the excitation strategy for each frequency band according to the local speech characteristics, resolving the contradiction between time resolution and periodicity representation.
2Reliability
If model-based speech coders use a high fidelity model to achieve high speech quality, then speech quality is improved, but the complexity of the system increases
Solution Approach 1:
The complex speech signal is segmented into multiple frequency bands, each processed by a relatively simple excitation model. This divides the overall system complexity into manageable parts while maintaining high fidelity through the combined effect of multiple bands, avoiding the need for a single complex high-fidelity model.
Solution Approach 2:
The system changes the representation parameters from a single complex time-domain model to multiple simpler frequency-band parameters. Each band uses simplified excitation parameters (noise level, periodicity) rather than attempting to model the entire speech signal with a single complex model, reducing overall system complexity while preserving quality.
3Productivity
If the bit rate is reduced to improve communication efficiency, then data transmission requirements are reduced, but speech quality and intelligibility deteriorate
Solution Approach 1:
By segmenting the speech signal into frequency bands and representing each with compact excitation parameters, the system achieves efficient compression. The segmented approach allows selective allocation of bits to different bands based on their importance, maintaining intelligibility while reducing overall bit rate requirements compared to uniform compression schemes.
Solution Approach 2:
The system transforms the speech signal into a compact parametric representation using frequency-band excitation parameters. This parameter change from time-domain waveforms to frequency-domain excitation characteristics enables much more efficient compression while preserving the essential features needed for speech intelligibility.
4Device complexity
If a single excitation source is used to represent all speech sounds, then the system is simpler to implement, but it cannot accurately represent both voiced and unvoiced sounds across different frequency bands
Solution Approach 1:
The speech representation is segmented into multiple frequency bands, each with its own excitation source. This segmentation allows the system to use simple excitation models in each band while collectively achieving accurate representation of both voiced and unvoiced sounds through the combination of multiple specialized excitation sources.
Data Source
AI summary
Encoding a sequence of digital speech samples into a bit stream includes dividing the digital speech samples into one or more frames, computing model parameters for a frame, and quantizing the model parameters to produce pitch bits conveying pitch information, voicing bits conveying voicing information, and gain bits conveying signal level information. One or more of the pitch bits are combined with one or more of the voicing bits and one or more of the gain bits to create a first parameter codeword that is encoded with an error control code to produce a first FEC codeword that is included in a bit stream for the frame. The process may be reversed to decode the bit stream.