Voiceprint Technology (III): voiceprint recognition technology

3.1　 Voiceprint recognition ： The core of voiceprint technology

3.1.1　 Name and concept

A broad sense , Voiceprint technology is a broad concept , It contains many different technologies and applications . Of all these technologies , Voiceprint recognition technology is the basis of other technologies . No matter it's the 5 The voiceprint segmentation and clustering technology will be introduced in chapter , Or no 6 This chapter will introduce the speech synthesis based on voiceprint 、 Voice separation and voice activity detection , Can not be separated from the cooperation with the voiceprint recognition model , The voiceprint recognition model can be pre trained （pre-trained）, It can also be joint training （joint training） Got . therefore , This chapter is also the most important part of this book 、 The core chapter .

Voiceprint recognition , Also known as speaker recognition , It corresponds to several kinds of expressions in English , for example voice recognition、speaker recognition、voiceprint recognition、talker recognition And so on are the same concept , That is to say, the pronunciation of different speakers , The technique of distinguishing according to the identity of the speaker .

Here we should pay attention to the combination of voiceprint recognition and speech recognition （speech recognition） Technology . Speech recognition is a technology that recognizes speech signal as text content , In most cases, I don't care who the speaker is , And need to be robust to different speakers' voices [1]. And voiceprint recognition technology —— Especially text independent voiceprint recognition technology —— On the contrary , It is necessary to identify the speaker's identity in different text content . From this we can see that , In a sense, speech recognition and voiceprint recognition are regarded as two mutual “ orthogonal ” The problem of ： Speech recognition hopes to filter out the information related to the identity of the speaker from the signal , Keep only text information ; Voiceprint recognition hopes to filter out the information related to the text from the signal , Only the identity information of the speaker is retained .

1.1 As mentioned in section , Everyone's hair