Joint factor analysis method and joint factor analysis voice-print verification method

Abstract

The invention discloses a joint factor analysis method, which comprises the following steps that: step 1: characteristics of an original voice is extracted; step 2: a universal background model is trained; step 3: an intrinsic space of a speaker is trained to obtain an intrinsic voice space of the speaker; step 4: a diagonal matrix model associated to the speaker is trained; and step 5: a plurality of channel spaces are respectively trained, and finally all channels are spliced to obtain a main channel space. The joint factor analysis method for training the speaker space in a serial way and training the channel space in a parallel way is adopted and is applied to a voice-print verification system, so the dependence of two low-dimensional spaces on the data during the training process can be greatly weakened, the processing time of the system can be greatly reduced, and the voice-print verification system can process more complicated environment and data. The invention also discloses a joint factor analysis voice-print verification method.

Description

technical field [0001] The invention relates to a voiceprint authentication method, in particular to a combined factor analysis method. The invention also relates to a combined factor analysis voiceprint authentication method. Background technique [0002] In all recognition tasks, the difference between the training environment and the test environment is the decisive factor affecting the performance of the voiceprint authentication system. In a complex real-world environment, the voiceprint authentication system needs to consider the influence of many factors, and the processing tasks are also very complex, including language differences, channel differences, voice recording microphone differences, and so on. [0003] The principle of the voiceprint authentication system is voiceprint recognition. Voiceprint recognition is a kind of biometric technology, which is a technology to automatically identify the identity of the speaker according to the voice parameters reflecti...

AI Technical Summary

Problems solved by technology

[0007] 1. The algorithm is complex and requires high training data;

[0008] 2. The assumptions about the independence of the speaker space and the channel space are not strict enough, so that the trained model cannot simulate the real situation;

[0009] 3. The processing of complex channels is not detailed enough to handle overly complex environments

Images