Directional noise suppression method based on assigned bandwidth

Abstract

The invention discloses a directional noise suppression method based on an assigned bandwidth. The directional noise suppression method is characterized in that step 1, a target voice channel transmission function can be estimated according to an expected evaluation value of a power spectrum of a transmission channel; step 2, a smooth noise component can be estimated and suppressed, and a voice blocking filter can be defined, and a target voice can be blocked, and the smooth noise component can be acquired, and then a target voice spectrum after the suppression of the smooth noise can be calculated; step 3, the directional noise can be suppressed, and the phase difference between the target voice and the observed voice can be determined, and when the observed voice of the current time frequency unit is the dominant directional noise, which can be filtered by using a band-pass filter; and finally, the target voice signal used for suppressing the smooth noise and the directional noise can be acquired; The directional noise suppression method based on the assigned bandwidth is advantageous in that the system is simple, and the difference between the directional noise and the smooth noise component can be effectively distinguished during the noise suppression process, and then based on the different natures, the suppression can be carried out by adopting different methods, and therefore the noise suppression effect is better, and the high applicability can be provided.

Description

technical field [0001] The invention relates to a noise suppression method for speech signals, in particular to a directional noise suppression method based on specified bandwidth. Background technique [0002] Regarding directional and non-directional noise: [0003] Intelligent voice-related signal acquisition equipment is usually disturbed by various noises while collecting user voice. According to the source of noise, it can be divided into two types: directional noise and non-directional noise. The so-called directional noise is generated by an independent sound source, and its direction is fixed relative to the microphone array of the acquisition terminal. In the field of signal processing, it is usually abstracted as a point source signal. For example, in the home environment, the sound of the TV, the sound of the alarm clock, etc. Non-directional noise refers to multiple noise sources with similar energies emitting noise at the same time, and these noise sources a...

AI Technical Summary

Problems solved by technology

[0009] 1) The existing technology relies on the sound source orientation information, assuming that the user's orientation is known, which affects the practicability, because in the actual use process, it is impossible to pre-fix the user's voice orientation

Since the direction information of the sound source relative to the microphone is based on the core clue information of the microphone array, most of the existing noise suppression technologies either use the sound source localization technology to estimate the direction information of the sound source, or limit the direction of the user.

In the field of dual-microphone noise reduction systems, due to the small number of microphones, it is difficult for the sound source localization algorithm to achieve high accuracy, which naturally affects the subsequent noise suppression effect

However, restricting the user's orientation, such as assuming that the user is 90 degrees in front of the microphone array, provides accurate orientation information, but significantly increases the freedom of the user's use, and the user experience is poor.

[0010] 2) The noise suppression model assumption in the prior art is too ideal, or based on a reverberation-free model, or the reverberation is very small

[0011] 3) In the existing technology, there are other technical means that can suppress the noise to a certain extent, but introduce too high voice distortion, which is not conducive to subsequent voice recognition

Therefore, many noise suppression techniques do not improve the accuracy of speech recognition, and even lead to a decrease in the recognition rate

Images