Airborne voice multi-dimensional noise reduction system and method based on artificial intelligence

Abstract

The invention discloses an airborne voice multi-dimensional noise reduction system and method based on artificial intelligence, and the system comprises a sound collection module, a noise detection module, a noise processing module, a central control system, a sound recognition module, and a noise reduction module. The noise processing module comprises a frequency calculation unit, a label adding unit, a noise sequencing unit and a noise extraction unit; the noise detection module detects a noise signal. According to the airborne voice multi-dimensional noise reduction system and method based on artificial intelligence disclosed by the invention, the noise frequencies of the detected multi-dimensional noise signals are sorted in a descending order, and in the process of performing noise reduction processing on the noise signal with the highest frequency, the noise signal with the lower frequency can be partially counteracted, so that the noise signal with the lower frequency is weakened, and the noise reduction efficiency is improved. The operation of performing noise reduction processing on each noise signal is avoided, and the step of performing noise reduction on the multi-dimensional noise signals one by one is reduced.

Description

Technical Field

[0001] The present invention relates to the field of speech processing technology, and in particular to an artificial intelligence-based airborne speech multi-dimensional noise reduction system and method. Background Art

[0002] In daily life, noise is one of the important factors that affect the quality of voice calls. Voice noise reduction technology improves voice quality and intelligibility by eliminating noise and extracting clean voice, which has great value in applications such as mobile communications, headsets, conference systems, and voice interaction. In real life or production scenarios, noise is divided into steady-state noise and non-steady-state noise. For example, environmental background noise and the noise of cars coming and going on the roadside are all steady-state noise; while mouse clicks, keyboard sounds, knocking sounds, air conditioning sounds, kitchen dishes colliding sounds, etc. are all non-steady-state noise.

[0003] Existing noise reduction processing methods mostly use one-to-one noise reduction. When multi-dimensional noise exists in the space at the same time, more resources and costs are required for noise reduction processing. In addition, the multi-dimensional noise reduction signals are independent of each other and have a certain influence on each other, which affects the noise reduction effect. Summary of the invention

[0004] The present invention discloses an airborne speech multi-dimensional noise reduction system and method based on artificial intelligence, aiming to solve the technical problems proposed in the background technology that one-to-one noise reduction is mostly adopted, and when multi-dimensional noise exists in the space at the same time, more resources and costs are needed to perform noise reduction processing, and the multi-dimensional noise reduction signals are independent of each other and have a certain influence on each other, which affects the noise reduction effect.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] An artificial intelligence-based airborne voice multi-dimensional noise reduction system comprises a sound collection module, a noise detection module, a noise processing module, a central control system, a sound recognition module and a noise reduction module, wherein the sound collection system is used to collect the original voice sound waves of the speaker, and the noise processing module comprises a frequency calculation unit, a label adding unit, a noise sorting unit and a noise extraction unit; the noise detection module detects noise signals and adds a label to each noise signal, wherein the label of each noise signal is unique, the frequency calculation unit is used to calculate the frequency of the noise signal detected by the noise detection module, the label adding unit is used to add a unique label to the noise signal detected by the noise detection module, the noise sorting unit sorts the noise signals according to the calculation results of the noise signal by the frequency calculation unit, and the noise extraction unit is used to extract the noise signal with the highest frequency.

[0007] In a preferred solution, the noise processing module performs corresponding fusion noise reduction processing on the labeled noise audio data features;

[0008] Performing Fourier transform on the noise audio data to obtain corresponding target frequency domain data;

[0009] Performing enhancement processing on the target frequency domain data to obtain corresponding frequency domain features;

[0010] Performing pooling processing on the frequency domain features to obtain corresponding enhanced features, and performing Fourier transform on the noise audio data to obtain corresponding target frequency domain data, including:

[0011] The noise audio data is divided by a time window function to obtain corresponding multi-dimensional time window audio data; the multi-dimensional time window audio data is parallel calculated by an analysis window function to obtain intermediate audio data corresponding to each of the multi-dimensional time window audio data; the obtained multi-dimensional intermediate audio data is parallel Fourier transformed to obtain sub-frequency domain data corresponding to each of the multi-dimensional intermediate audio data; the sub-frequency domain data corresponding to each of the multi-dimensional intermediate audio data are jointly determined as target frequency domain data corresponding to the noise audio data. The noise processing module includes a noise reduction circuit and a noise reduction speaker; the noise reduction circuit is used to generate an electrical signal with the same amplitude and opposite phase as the noise signal according to the frequency of the noise signal, and the noise reduction speaker is used to convert the electrical signal generated by the noise reduction circuit into a sound wave to perform noise reduction processing on the noise signal.

[0012] In a preferred embodiment, the sound recognition module includes a noise analysis unit, a noise removal unit and a storage database, the noise analysis unit is used to analyze the noise signal data, the central control system judges whether the noise signal is a human voice noise signal based on the analysis result of the noise analysis unit and the storage data in the storage database, the noise removal unit is used to remove the human voice noise signal judged by the central control system from the noise signal detected by the noise detection module, and the storage database is used to store the removed human voice noise signal data.

[0013] In a preferred solution, the noise reduction module includes a noise reduction model library and a data update unit. The noise reduction model library is used to store the iconic noise signals in different scenes and times to form a noise reduction model. The noise reduction model library also includes a collection unit and a classification unit. The collection unit is used to collect the iconic noise signals of each environmental point. The classification unit is used to slice the iconic noise signals into uniform time slices and classify them according to the collection time and scene. The output end of the noise analysis unit and the storage database are connected to the input end of the central control system, the output end of the central control system is connected to the input end of the noise removal unit, and the output end of the noise removal unit is connected to the input end of the storage database.

[0014] An artificial intelligence-based multi-dimensional noise reduction method for airborne speech comprises the following steps:

[0015] S1, the sound collection module obtains the target voice signal;

[0016] S2, transmitting the target speech to the noise detection module for detection, processing the detected noise, and determining the noise type;

[0017] S3, according to the classification result of the noise, the noise removal unit removes the sound of conversation;

[0018] S4, processing the remaining noise and sorting them in descending order;

[0019] S5, performing noise reduction processing on the first noise of the descending order;

[0020] S6. Repeat S1-S5 to reduce the noise in the second and third order in descending order respectively, until the noise in multiple dimensions in the environment is reduced.

[0021] As can be seen from the above, an airborne voice multi-dimensional noise reduction system based on artificial intelligence includes a sound collection module, a noise detection module, a noise processing module, a central control system, a sound recognition module and a noise reduction module. The sound collection system is used to collect the original voice sound waves of the speaker, and the noise processing module includes a frequency calculation unit, a label adding unit, a noise sorting unit and a noise extraction unit; the noise detection module detects the noise signal and adds a label to each noise signal, and the label of each noise signal is unique. The frequency calculation unit is used to calculate the frequency of the noise signal detected by the noise detection module, and the label adding unit is used to add a unique label to the noise signal detected by the noise detection module. The noise sorting unit sorts the noise signal according to the calculation result of the noise signal by the frequency calculation unit, and the noise extraction unit is used to extract the noise signal with the highest frequency. The airborne voice multi-dimensional noise reduction system and method based on artificial intelligence provided by the present invention have the following technical effects:

[0022] 1: Sort the detected multi-dimensional noise signals in descending order of noise frequency, and first perform noise reduction on the noise with the highest frequency in the descending order. In the process of noise reduction on the noise signal with the highest frequency, the noise signal with lower frequency will be partially offset, weakening the noise signal with lower frequency. Then, the remaining multi-dimensional noise signals are processed in sequence, avoiding the need to perform noise reduction on each noise signal, reducing the steps of performing noise reduction on multi-dimensional noise signals one by one, and avoiding the mutual influence between multi-dimensional noise reduction signals, making the process of multi-point noise reduction more concise and the processing of multi-point noise reduction more energy-efficient.

[0023] 2: The human voice noise signal is removed from the multi-dimensional noise signal through the sound elimination unit, which avoids the multi-point noise reduction system from reducing the human voice noise and affecting people's normal conversation. The multi-point noise reduction system can achieve differentiated noise reduction for the noise signal, making the multi-point noise reduction system more intelligent and humane.

[0024] 3: By obtaining the noise source in the user's current environment, targeted speech noise reduction can be performed, thereby improving the efficiency of speech noise reduction.

[0025] 4: Process all audio data and obtain a corresponding enhanced feature for each noisy audio data; fuse the enhanced features corresponding to all the noisy audio data with the original audio label to obtain a corresponding original audio feature, and fuse the enhanced features corresponding to all the noisy audio data with the noisy audio label to obtain a corresponding original audio feature, which can effectively improve the noise reduction effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is an overall flow chart of an artificial intelligence-based airborne speech multi-dimensional noise reduction method proposed by the present invention. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0028] Reference Figure 1 , an airborne voice multi-dimensional noise reduction system based on artificial intelligence, comprising a sound collection module, a noise detection module, a noise processing module, a central control system, a sound recognition module and a noise reduction module, wherein the sound collection system is used to collect the original voice sound waves of the speaker, and the noise processing module comprises a frequency calculation unit, a label adding unit, a noise sorting unit and a noise extraction unit; the noise detection module detects noise signals and adds a label to each noise signal, and the label of each noise signal is unique, the frequency calculation unit is used to calculate the frequency of the noise signal detected by the noise detection module, the label adding unit is used to add a unique label to the noise signal detected by the noise detection module, the noise sorting unit sorts the noise signals according to the calculation results of the noise signal by the frequency calculation unit, and the noise extraction unit is used to extract the noise signal with the highest frequency.

[0029] In a preferred embodiment, the noise processing module performs corresponding fusion noise reduction processing on the labeled noise audio data features;

[0030] Performing Fourier transform on the noise audio data to obtain corresponding target frequency domain data;

[0031] Performing enhancement processing on the target frequency domain data to obtain corresponding frequency domain features;

[0032] Pooling is performed on the frequency domain features to obtain corresponding enhanced features.

[0033] In a preferred embodiment, Fourier transform is performed on the noise audio data to obtain corresponding target frequency domain data, including:

[0034] The noise audio data is divided by a time window function to obtain corresponding multi-dimensional time window audio data; the multi-dimensional time window audio data is calculated in parallel by an analysis window function to obtain intermediate audio data corresponding to each of the multi-dimensional time window audio data; the obtained multi-dimensional intermediate audio data is subjected to parallel Fourier transform to obtain sub-frequency domain data corresponding to each of the multi-dimensional intermediate audio data; the sub-frequency domain data corresponding to each of the multi-dimensional intermediate audio data are jointly determined as the target frequency domain data corresponding to the noise audio data.

[0035] In a preferred embodiment, the noise processing module includes a noise reduction circuit and a noise reduction speaker; the noise reduction circuit is used to generate an electrical signal with the same amplitude and opposite phase as the noise signal according to the frequency of the noise signal, and the noise reduction speaker is used to convert the electrical signal generated by the noise reduction circuit into a sound wave to perform noise reduction processing on the noise signal.

[0036] In a preferred embodiment, the sound recognition module includes a noise analysis unit, a noise removal unit and a storage database, the noise analysis unit is used to analyze the noise signal data, the central control system judges whether the noise signal is a human voice noise signal based on the analysis result of the noise analysis unit and the storage data in the storage database, the noise removal unit is used to remove the human voice noise signal judged by the central control system from the noise signal detected by the noise detection module, and the storage database is used to store the removed human voice noise signal data.

[0037] In a preferred embodiment, the noise reduction module includes a noise reduction model library and a data update unit. The noise reduction model library is used to store the iconic noise signals in different scenes and times to form a noise reduction model. The noise reduction model library also includes a collection unit and a classification unit. The collection unit is used to collect the iconic noise signals of each environmental point. The classification unit is used to cut the iconic noise signals into uniform time-length slices and classify them according to the collection time and scene.

[0038] In a preferred embodiment, the outputs of the noise analysis unit and the storage database are connected to the input of the central control system, the output of the central control system is connected to the input of the noise removal unit, and the output of the noise removal unit is connected to the input of the storage database.

[0039] An artificial intelligence-based multi-dimensional noise reduction method for airborne speech comprises the following steps:

[0040] S1, the sound collection module obtains the target voice signal;

[0041] S2, transmitting the target speech to the noise detection module for detection, processing the detected noise, and determining the noise type;

[0042] S3, according to the classification result of the noise, the noise removal unit removes the sound of conversation;

[0043] S4, processing the remaining noise and sorting them in descending order;

[0044] S5, performing noise reduction processing on the first noise of the descending order;

[0045] S6. Repeat S1-S5 to reduce the noise in the second and third order in descending order respectively, until the noise in multiple dimensions in the environment is reduced.

[0046] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An artificial intelligence-based airborne voice multi-dimensional noise reduction system, comprising a sound collection module, a noise detection module, a noise processing module, a central control system, a sound recognition module and a noise reduction module, characterized in that: The sound collection system is used to collect the original voice sound waves of the speaker. The noise processing module includes a frequency calculation unit, a label adding unit, a noise sorting unit and a noise extraction unit. The noise detection module detects the noise signal and adds a label to each noise signal. The label of each noise signal is unique. The frequency calculation unit is used to calculate the frequency of the noise signal detected by the noise detection module. The label adding unit is used to add a unique label to the noise signal detected by the noise detection module. The noise sorting unit sorts the noise signal according to the calculation result of the noise signal by the frequency calculation unit. The noise extraction unit is used to extract the noise signal with the highest frequency.

2. The multi-dimensional noise reduction system for airborne speech based on artificial intelligence according to claim 1, characterized in that: The noise processing module performs corresponding fusion noise reduction processing on the labeled noise audio data features; Performing Fourier transform on the noise audio data to obtain corresponding target frequency domain data; Performing enhancement processing on the target frequency domain data to obtain corresponding frequency domain features; Pooling is performed on the frequency domain features to obtain corresponding enhanced features.

3. The multi-dimensional noise reduction system for airborne speech based on artificial intelligence according to claim 2, characterized in that: Performing Fourier transform on the noise audio data to obtain corresponding target frequency domain data, including: The noise audio data is divided by a time window function to obtain corresponding multi-dimensional time window audio data; the multi-dimensional time window audio data is calculated in parallel by an analysis window function to obtain intermediate audio data corresponding to each of the multi-dimensional time window audio data; the obtained multi-dimensional intermediate audio data is subjected to parallel Fourier transform to obtain sub-frequency domain data corresponding to each of the multi-dimensional intermediate audio data; the sub-frequency domain data corresponding to each of the multi-dimensional intermediate audio data are jointly determined as the target frequency domain data corresponding to the noise audio data.

4. The multi-dimensional noise reduction system for airborne speech based on artificial intelligence according to claim 1, characterized in that: The noise processing module includes a noise reduction circuit and a noise reduction speaker; the noise reduction circuit is used to generate an electrical signal with the same amplitude and opposite phase as the noise signal according to the frequency of the noise signal, and the noise reduction speaker is used to convert the electrical signal generated by the noise reduction circuit into a sound wave to perform noise reduction processing on the noise signal.

5. The multi-dimensional noise reduction system for airborne speech based on artificial intelligence according to claim 1, characterized in that: The sound recognition module includes a noise analysis unit, a noise removal unit and a storage database. The noise analysis unit is used to analyze the noise signal data. The central control system determines whether the noise signal is a human voice noise signal based on the analysis result of the noise analysis unit and the storage data in the storage database. The noise removal unit is used to remove the human voice noise signal determined by the central control system from the noise signal detected by the noise detection module. The storage database is used to store the removed human voice noise signal data.

6. The multi-dimensional noise reduction system for airborne speech based on artificial intelligence according to claim 1, characterized in that: The noise reduction module includes a noise reduction model library and a data update unit. The noise reduction model library is used to store iconic noise signals in different scenes and times to form a noise reduction model. The noise reduction model library also includes a collection unit and a classification unit. The collection unit is used to collect iconic noise signals at each environmental point. The classification unit is used to cut the iconic noise signals into uniform time-length slices and classify them according to the collection time and scene.

7. The artificial intelligence-based airborne voice multi-dimensional noise reduction system according to claim 6, characterized in that: The output ends of the noise analysis unit and the storage database are connected to the input end of the central control system, the output end of the central control system is connected to the input end of the noise removal unit, and the output end of the noise removal unit is connected to the input end of the storage database.

8. An artificial intelligence-based multi-dimensional noise reduction method for airborne speech, applied to an artificial intelligence-based multi-dimensional noise reduction system for airborne speech according to any one of claims 1 to 7, characterized in that: The specific steps include: S1, the sound collection module obtains the target voice signal; S2, transmitting the target speech to the noise detection module for detection, processing the detected noise, and determining the noise type; S3, according to the classification result of the noise, the noise removal unit removes the sound of conversation; S4, processing the remaining noise and sorting them in descending order; S5, performing noise reduction processing on the first noise of the descending order; S6. Repeat S1-S5 to reduce the noise in the second and third order in descending order respectively, until the noise in multiple dimensions in the environment is reduced.

Images