Research On The Performance Of Speech Features In Gender-based Speaker Recognition

Speaker recognition,as its name suggests,is a technique that uses the acoustic characteristics of a speaker to perform authentication,also known as voiceprint recognition.Nowadays it has been widely used in speech,which is the most widely used and convenient tool in people's daily life.Human voiceprint features are also unique,stable,easy to use,not easy to lose and universal,so the speaker recognition technology uses the characteristics of human voiceprint features,and is also optimistic in the field of identity recognition.In recent years,the application fields of voiceprint recognition technology have been increasing at home and abroad.For example,in the public security field,the speaker recognition system,voiceprint recognition family phone system,voiceprint recognition and location tracking system,residential buildings,voice registration of important office occasions,online transaction identification and many other applications.But nowadays hackers or real-world analog piracy technologies are very common,which is very urgent to support the accuracy and stability of the voiceprint recognition system.In terms of the current research status in the field of speaker recognition,the speaker recognition technology is roughly composed of sound text collection,sound text preprocessing,speech feature extraction,sound text modeling and recognition.In the research,we are eager to find an algorithm that can make the recognition rate higher.With the continuous development of artificial intelligence technology,the method of using deep learning has already brought the research to a new height,and it is too difficult to improve the recognition efficiency through the algorithm direction.In other words,the optimization of the recognition algorithm has reached a bottleneck period.In the pursuit of high accuracy,we only pay attention to the improvement of the accuracy of the algorithm,but ignore the direction of modeling the speech text in the speaker extraction process.Today we are using the feature extraction of the overall speaker and the establishment of a feature model.This paper proposes a concept of refinement modeling.Of course,there are many ways to refine modeling,such as sub-regional,age-specific,and gender-specific.The main point of this paper is to distinguish between speaking people by gender.Feature extraction and acoustic feature models are performed separately.A speaker recognition method based on vector quantization and a speaker recognition method based on deep learning are used.Features such as MFCC,LPCC,MFSC,and dual MFCC are extracted for male and female voice features.Explore the genders to which the phonetic features apply.Through the analysis and comparison of the experimental results,the results show that in the case of using the same recognition system,the MFCC and its related feature extraction methods are more capable of representing male voices than female voices.LPCC's ability to represent female voices is better than that of male voices.And when using high-latitude recognition models for experiments,the performance of MFSC is higher than MFCC.This provides a direction for future research on improving the performance of speaker recognition systems,that is,male and female speakers can be studied separately,and algorithms or models applicable to different genders can be separately explored to fundamentally improve system performance.