CN102110441A - Method for generating sound masking signal based on time reversal - Google Patents

Abstract

The invention relates to a method for generating a sound masking signal based on time reversal. The method obtains a corresponding sound masking signal according to a target sound source signal. The sound masking signal has a long-term amplitude spectrum similar to that of the target sound source signal, and speech can The understanding is very low; the specific steps of the method include: using a microphone or a microphone array to pick up the target sound source signal, and obtaining a clean target sound source signal through preprocessing; y(t)=x(-t) performs time inversion on each frame signal in the time domain to obtain the corresponding sound masking signal of the target sound source; wherein, x(t) represents a frame signal, and y(t) represents the frame output signal. The time-reversed sound masking signal of the present invention has a similar long-term amplitude spectrum to the target sound source signal, and has an advantage in energy masking, and the time-reversed sound masking signal is completely incomprehensible or partially intelligible, and will not become a new interference Sound source.

Description

A Time-Reversal-Based Acoustic Masking Signal Generation Method

technical field

The invention relates to a method for generating a sound masking signal, in particular to a method for generating a sound masking signal based on time reversal.

Background technique

Studies have shown that speech is the most disturbing sound signal in the room, and the speech intelligibility of speech signals will reduce work efficiency. In some occasions, due to the high speech intelligibility, the privacy of personal calls cannot be guaranteed, and there is an urgent need for a method that can reduce the speech intelligibility and protect the privacy of personal calls. Sound masking refers to the method of adding a natural or artificially synthesized sound into the environment, covering the sound of the target sound source through auditory masking, and reducing the intelligibility of the target sound source signal. At present, acoustic masking technology based on acoustic masking signals is considered to be one of the necessary measures to improve the acoustic environment of open offices. In the existing sound masking signal generation methods, their masking signal sources are generally noise signals, such as: white noise, pink noise, air-conditioning noise and various noises artificially generated, but the noise signal has no correlation with the target sound source signal. The masking efficiency is very low, which requires the noise energy at the receiving point to be much greater than the signal energy of the target sound source in order to reduce the speech intelligibility of the target sound source. The problem is that too much noise energy can add to the annoyance of the sound, making it unbearable. If the acoustic masking signal is generated by processing the target sound source signal, the two have correlation, which can improve the masking efficiency. Therefore, it is necessary to find a more effective sound masking signal, which is obtained by processing the target sound source signal, and is superior to the noise masking signal in terms of masking efficiency.

Contents of the invention

The purpose of the present invention is to propose a time-reversal-based sound masking signal generation method, so that the sound masking signal is generated by processing the target sound source signal, and the two have correlation, which can improve the masking efficiency.

In order to achieve the above object, the present invention provides a method for generating a sound masking signal based on time inversion, the method obtains a corresponding sound masking signal according to the target sound source signal, and the sound masking signal has a long-term amplitude similar to the target sound source signal Spectrum, and speech intelligibility is very low; the specific steps of the method include:

Step 1): Use a microphone or a microphone array to pick up the target sound source signal, and obtain a clean target sound source signal through preprocessing;

Step 2): The target sound source signal obtained according to the step 1) is divided into frames according to the specific time window length, and the sound source of the target sound source is obtained after the time inversion of each frame signal in the time domain according to formula (1). masking signal;

y(t)=x(-t) (1)

Wherein, x(t) represents a frame signal, and y(t) represents an output signal of the frame.

The preprocessing in step 1) includes: speech enhancement, noise reduction, sound source localization and sound feature recognition.

In the step 2), the frame length is 150 ms-500 ms when the frame is divided according to the specific time window length. The step 2) further includes: adding a time window function smoothing to each time-reversed signal. The edge decay of the time window function is fast and the endpoint is 0.

The advantage of the present invention is that the time-reversed sound masking signal of the present invention has a similar long-term amplitude spectrum to the target sound source signal, and has an advantage in energy masking, and the time-reversed sound masking signal is completely incomprehensible or partially intelligible, It will not become a new source of disturbing sound. The masking performance of the time-reversed sound masking signal is much higher than that of the noise masking signal. Under the same energy ratio of the target sound source signal to the sound masking signal, it can greatly reduce speech intelligibility and protect call privacy. In addition, the time-reversed sound masking signal generation method has a simple signal processing process and is suitable for real-time processing.

Description of drawings

Fig. 1 is a kind of flow chart of the generation method of sound masking signal based on time reversal of the present invention;

Fig. 2 is the oscillogram after the target sound source signal of the present invention is preprocessed and carried out framing;

Figure 3 is a waveform diagram after time reversal based on the waveform in Figure 2;

Fig. 4 is based on the waveform in Fig. 3 and adds the sound masking signal wave diagram after Tukey window function smoothing;

FIG. 5 is a masking waveform diagram obtained by superimposing the target sound source signal in FIG. 2 and the sound masking signal in FIG. 4 .

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

The masking signal has a long-term amplitude spectrum similar to the target sound source signal, and the speech intelligibility is very low. The sound masking signal produced by the method of the present invention is far superior to the existing noise masking signal in terms of masking efficiency, and is shown in the same target sound source signal and sound masking signal energy ratio TMR (Target-to-Masker ratio) situation, so that The speech intelligibility of the target sound source signal is lower, which effectively protects the privacy of the call.

A flowchart of a method for generating a sound masking signal based on time reversal in the present invention is shown in FIG. 1 . Technical scheme of the present invention comprises the steps:

Step 1): A single microphone, multiple microphones or a microphone array picks up the target sound source signal and performs preprocessing. The preprocessing process includes speech enhancement, noise reduction, etc.

Step 2): Framing the preprocessed clean target sound source signal according to a specific time window length, performing time inversion of each frame signal in the time domain, and adding a time window function to each frame signal after time inversion, The sound masking signal of the target sound source is obtained between smoothing frames before and after. The sound masking signal is replayed through the speaker system, which has a masking effect on the target sound source signal, reduces the speech intelligibility of the target sound source signal, and protects the call privacy of the target sound source

Below each step of the present invention is described in further detail in conjunction with Fig. 2, Fig. 3, Fig. 4 and Fig. 5:

In the step 1, the specific implementation is as follows:

Pick up the target sound source signal through a single microphone, multiple microphones or a microphone array. Due to the existence of background noise and the difference in the position of the target sound source and the microphone, the picked up signal may contain noise and other signals. A clean target can be obtained through preprocessing For the sound source signal, the preprocessing process may include speech enhancement, noise reduction, sound source localization, sound source feature recognition, etc.

In the step 2, the specific implementation is as follows:

The waveform diagram of the target sound source signal of the present invention after being preprocessed and divided into frames, as shown in Figure 2, the preprocessed signal is divided into frames according to a specific time window length, and the recommended range of frames is between 150ms and 500ms. The example in the figure is 200ms. Then, as shown in FIG. 3 , time inversion processing is performed on the framed signal, and each frame signal is inverted in the time domain to obtain an inverted signal. At this time, the signal is discontinuous at the frame-to-frame connection, and is smoothed by a window function, and the smoothed signal is shown in FIG. 4 .

For time-reversal signals, the speech intelligibility will change as the frame length changes. Specifically, when the frame length is less than 50ms, the time-reversal signal is almost completely intelligible. As the frame length increases, the speech intelligibility decreases. When the frame length is 130ms, it is about 50% intelligible, until the frame length is close to 200ms, it is completely incomprehensible. In this invention, it is hoped to design a masking signal that is similar to the long-term amplitude spectrum of the target sound source signal, but completely incomprehensible or partially intelligible, so the frame length is recommended to be more than 150ms. If the frame length is too short, the sound masking signal itself is intelligible , cannot be masked. Theoretically, the frame length is more than 200ms, and the time reversal signal is completely incomprehensible. However, considering the requirements of real-time processing in the application, the frame length range is recommended to be below 500ms.

The time-reversed signal is shown in Figure 3. There is discontinuity between frames, which may generate noise and increase the annoyance of the time-reversed masking signal. Here, the time window function is used for smoothing. For the selection of the window function, it is required that the attenuation at the edge of the window function is fast and the end point is 0, and most of the waveform information of the time-reversed sound masking signal is retained as much as possible. Figure 4 shows the signal waveform with a frame end point value of 0 after being smoothed by the window function picture.

The sound masking signal of the target sound source in the legend is shown in Figure 4, and it is replayed through the speaker system. The speaker system can be a single speaker, multiple speakers or a speaker array, depending on the system space design structure and the desired masking effect , to adjust the amplitude of the input signal of the loudspeaker to realize the speech intelligibility control of the target sound source signal. In general, the energy ratio of the target sound source signal to the time-reversed sound masking signal at the receiving point is between -5dB and 0dB, and the speech intelligibility will decline to a certain extent. If you want lower speech intelligibility, The energy of the time-reversed sound masking signal can be appropriately increased.

Specific examples of the present invention are as follows:

The signal processing process of this embodiment is simulated in MATLAB software, and the speech intelligibility is evaluated through auditory subjective experiments on the processing results. Assuming that a clean target sound source signal has been obtained, as shown in Figure 2, then the signal in Figure 2 is divided into frames with 200ms as a frame. Inversion processing, as shown in Figure 3, at this time, there is discontinuity between frames, in order to eliminate the discontinuity, Tukey window function is added for smoothing, and the window function MATLAB expression is "tukeywin(L, 0.2)", where L is the number of sampling points in the frame, and finally the time-reversed sound masking signal of the target sound source signal is obtained, as shown in Figure 4. After adjusting the energy ratio of the target sound source signal Figure 2 and the sound masking signal Figure 4, the output test signal is obtained after superposition as shown in Figure 5, where Figure 5 shows the result when the energy ratio is 0dB. The speech intelligibility error rate output in Figure 5 is tested, and the noise signal is used as a reference. The experimental results are shown in Table 1, which shows that the time-reversed sound masking signal has a higher performance than the noise masking signal under the same energy ratio. Good ability to reduce speech intelligibility, for example, when the energy ratio is -10dB, the intelligibility error rate is 97% after the time-reversed sound masking signal masks the target sound source signal, there is no speech intelligibility, and the noise masking signal masks There is little loss of speech intelligibility after the target source signal.

Table 1 Experimental results - error rate of speech intelligibility

Energy ratio TMR -15dB -10dB -5dB 0dB Time-reversed acoustic masking signal 96% 97% 31% 8% noise masking signal 12.5% 0.13% 0.2% 0.1%

In this embodiment, although 200 ms is used as the frame length and Tukey is used as the time window, this is only an illustration of the method provided by the present invention, and the above embodiment is only used to illustrate the technical solution of the present invention rather than limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent replacements to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all of them should be included in the scope of the present invention. within the scope of the claims.

Claims (6)

1. A method for generating a sound masking signal based on time reversal, the method obtains a corresponding sound masking signal according to a target sound source signal, the sound masking signal has a long-term amplitude spectrum similar to the target sound source signal, and the speech is intelligible The degree is very low; the specific steps of the method include: Step 1): Use a microphone or a microphone array to pick up the target sound source signal, and obtain a clean target sound source signal through preprocessing; Step 2): The target sound source signal obtained according to the step 1) is divided into frames according to the length of a specific time window, and the time inversion of the time domain is performed on each frame signal according to formula (1) to obtain the corresponding sound masking signal; y(t)=x(-t) (1) Wherein, x(t) represents a frame signal, and y(t) represents an output signal of the frame. 2. The method for generating a sound masking signal based on time inversion according to claim 1, wherein the preprocessing in step 1) includes: speech enhancement, noise reduction, sound source localization and sound feature recognition. 3. The sound masking signal generation method based on time inversion according to claim 1, characterized in that, in the step 2), the frame length is 150 ms-500 ms when the frame is divided according to the specific time window length. 4. The sound masking signal generation method based on time inversion according to claim 1, characterized in that, said step 2) further comprises: adding a time window function smoothing to each frame signal of time inversion. 5. The method for generating a sound masking signal based on time inversion according to claim 4, characterized in that, the edge of the time window function decays quickly and the endpoint is 0. 6. The method for generating a sound masking signal based on time inversion according to claim 4 or 5, wherein the time window function is a Tukey window function.

Images