Key Technology Research On Audio Information Hiding And Information Security Application For Speech Recognition

Digital products broadcasted with exponential growth have increased large amount with the development of Internet and multimedia since 1990's. This lets the ancient steganography get chance to have a carrier for new life, and it forms a new research field called information hiding. Information hiding is now an important focus of information security. Since digital audio, especially digital music and speech communication, is closer to people's lives, its information hiding has good promise for application. Audio information hiding can not only be used in secret communication of espionage or confidential department, but also be used in civil purposes such as personal privacy protection, security use of Internet and right protection for digital product, so its research has practicability, economy value and the meaning of country security.Speech is an important branch of audio. Inevitably we should have cross with speech recognition, because we must combine the research of speech character while studying the information hiding of speech. Methods and results of speech recognition technology can be combined to information hiding for the application purposes of information security. It is of great importance for implement of economy value to do research on information secutity application of speech recognition and information hiding of speech, and to do research on their application fields and scenes of application.This paper looks at the process of speech conversation from the viewpoint of information hiding, and discovers the analogy relation between speech recognition and audio information hiding. This paper studies on audio information hiding and combination research with speech recognition, and gives the following creative results.1. An information hiding method using the redundancy after the endpoint of Chinese speech is provided. The phoneme at the end of syllable is always voiced speech in Chinese, while voiced speech can be regarded as an output for quasi-periodic sequence of pulses acting on vocal tract. This property of Chinese speech can be used in endpoint detection to distinguish "sound or no sound". By using this endpoint detection method, periodic redundancy of speech in time domain can be decided, and hiding in the redundancy is fullfiled.2. An information hiding method using MFCC is provided. MFCC is the main parameter for speech recognition. In order to hide in MFCC, this paper gives answers to the following questions: (1). Criterion for MFCC selection. (2). Solution for getting log energe from changed MFCC. (3). Solution for reverse transformation of Mel frequency filter bank. Based on these answers, we can hide data in MFCC successfully.3. An information hiding method in AAC is provided. In the calculation test step of AAC, codebook selection can be used for hiding as bit 0 or 1, considering the possibility of same length of the shortest coded bits of sfb quantized frequent value with different codebook.4. Chinese speech verification code is constructed because of the short time character of Chinese pronunciation and the research result that the performance will decrease sharply in noise surround inevitably for most speech recognition system. It solves the problem of the adaption between WEB application and synthesis speed mainly, and be an optional safety solution for common user's logging on Internet bank. It is an important information security application for speech recognition results.5. A sample application combining audio watermarking with speech recognition is carried out. in the process of automatic speech service, audio watermarking can be embedded in automatic speech, and customer's speech terminal can call speech recognition engine by watermarking detection and assuring of automatic speech, thus customer's speech can interact with automatic speech.Audio information hiding research has wide space to explore nowdays, especially in formatted audio media hiding, hiding combinated to speech recognition and hiding combinated to low bit rate speech coding. Moreover we should enhence field application and integrated application research for speech recognition and audio information hiding.