Howling detection in an audio system

Abstract

An audio system (20) includes one or more microphones (23, 24), a speaker (28), and processing circuitry (26). The one or more microphones are configured to output electrical signals in response to sound waves incident on the microphones. The processing circuitry is configured to amplify and filter the electrical signals using a digital filter (42) in order to generate a drive signal for input to the speaker, to detect that a howling signal is being amplified by identifying a plurality of frequency components in the drive signal that are spaced apart from each other by a constant frequency difference, and to adjust the drive signal to suppress the howling signal.

Description

Technical Field

[0001] The embodiments described herein generally relate to audio systems, and more particularly to methods and systems for detecting and suppressing feedback in audio systems such as hearing aids and sound enhancement systems.

[0002] background Some audio systems include one or more microphones and one or more speakers. In various situations, acoustic feedback from the speaker to the microphone can cause a degrade in audio quality or even lead to unwanted howling.

[0003] Methods for handling acoustic feedback are known in the art. For example, in a review paper entitled “Fifty Years of Acoustic Feedback Control: State of the Art and Future Challenges” (Proceeding of the IEEE, Vol. 99, No. 2, February 2011), the authors aimed to provide an overview of the latest technologies in the field of acoustic feedback control.

[0004] In a paper titled “Temporal Howling Detector for Speech Reinforcement Systems” (published in Acoustics, Vol. 4, No. 4, pp. 967-995, November 2022), the authors discuss the problem of howling detection in speech enhancement systems for use in howling control mechanisms.

[0005] Overview The embodiments described herein provide an audio system including one or more microphones, a speaker, and processing circuitry. The one or more microphones are configured to output an electrical signal in response to sound waves incident on the microphone. The processing circuitry is configured to amplify and filter the electrical signal using a digital filter to generate a drive signal for input to the speaker, detect increasing feedback by identifying multiple frequency components in the drive signal that are spaced at constant frequency differences from each other, and adjust the drive signal to suppress the feedback signal.

[0006] In some embodiments, the frequency difference is inversely proportional to the system's loop delay. In other embodiments, the processing circuitry is configured to identify the dominant frequency component among a plurality of frequency components and suppress the howling signal by reducing the gain of at least the dominant frequency component. In still other embodiments, the processing circuitry is configured to gradually increase the gain of the dominant frequency component after reducing the gain, while verifying that the drive signal does not contain multiple frequency components.

[0007] In one embodiment, the digital filter includes multiple sub-filters with corresponding frequency bands and gains, and the processing circuitry is configured to suppress the howling signal by reducing the gain of the sub-filters whose frequency bands contain at least a dominant frequency component. In another embodiment, the processing circuitry is configured to identify that the drive signal contains multiple frequency components by: (i) calculating a normalized autocorrelation signal within a specified hysteresis range based on the drive signal; (ii) calculating the maximum value of the normalized autocorrelation signal within the hysteresis range; and (iii) detecting that the maximum value exceeds a predefined threshold. In yet another embodiment, the processing circuitry is configured to transform the normalized autocorrelation signal to the frequency domain and estimate the howling frequency of the howling signal by identifying the dominant frequency component in the frequency domain.

[0008] In some embodiments, the processing circuitry is configured to estimate the loop delay by finding a hysteresis value that maximizes the normalized autocorrelation signal within a hysteresis range. In other embodiments, the digital filter includes a beamforming filter.

[0009] According to the embodiments described herein, a method for audio processing is also provided, comprising: installing one or more microphones that output electrical signals in response to sound waves incident on the microphones; installing a loudspeaker; amplifying and filtering the electrical signals using a digital filter to generate a drive signal for input to the loudspeaker; detecting an increasing feedback signal by identifying multiple frequency components in the drive signal that are spaced at constant frequency differences from each other; and adjusting the drive signal to suppress the feedback signal.

[0010] According to another embodiment described herein, a head-mounted device (HMD) is also provided, comprising a frame, one or more microphones, a speaker, and processing circuitry. The frame is configured to be mounted on the head of a subject. One or more microphones are mounted on the frame and configured to output electrical signals in response to sound waves incident on the microphones. The speaker is mounted on the frame. The processing circuitry is configured to amplify and filter the electrical signals using a digital filter to generate a drive signal for input to the speaker, detect an increasing howling signal by identifying multiple frequency components in the drive signal that are spaced at constant frequency differences from each other, and adjust the drive signal to suppress the howling signal.

[0011] In some embodiments, the HMD includes devices selected from a list including: eye-wearing devices, eyeglasses, eyeglass frames, goggles, helmets, face shields, headbands, and clip-on devices. In other embodiments, one or more microphones are mounted on the front of the frame, and speakers are mounted on the frame near the subject's ears.

[0012] These and other embodiments will be more fully understood from the following detailed description of the embodiments taken in conjunction with the accompanying drawings, wherein: Attached Figure Description

[0013] Figure 1 This is a schematic illustration of a hearing aid device integrated into an eyeglass frame according to embodiments described herein; Figure 2 This is a block diagram schematically illustrating details of the processing circuitry in a hearing aid device according to an embodiment described herein; Figure 3A and Figure 3B This is a schematic illustration of an example howling signal driving a loudspeaker according to an embodiment described herein, and a corresponding frequency density spectrum diagram; Figure 4 This is a schematic flowchart illustrating a method for howling detection and suppression according to embodiments described herein; Figure 5 This is a flowchart schematically illustrating a method for detecting and suppressing howling signals based on computational time-domain autocorrelation according to embodiments described herein; Figure 6 This is a flowchart schematically illustrating a method for activating / deactivating howling suppression according to embodiments described herein; Figure 7 The embodiments described herein are illustrated schematically. Figure 3A The diagram shows the howling signal and the time frame used to calculate the normalized autocorrelation signal. Figure 8 The embodiments described herein are illustrated schematically. Figure 7 A diagram illustrating the normalized autocorrelation signal of the howling signal; and Figure 9 This schematically illustrates the transformation to the frequency domain according to the embodiments described herein. Figure 8 A diagram illustrating the normalized autocorrelation signal.

[0014] Detailed description of the embodiments Overview Various audio systems, such as hearing aids and sound enhancement systems, include one or more microphones and one or more speakers. The signal output from the microphones is typically processed to generate the audio output for the speakers. Such processing should reproduce the natural auditory experience as closely as possible while minimizing bothersome artifacts.

[0015] One type of artifact is the loud howling that can result from acoustic feedback from the speaker's audio output to the microphone's input. This howling occurs when the acoustic feedback gain of the hearing aid at a given frequency exceeds a certain threshold (e.g., a gain greater than 1). Conventional solutions to this problem include reducing the gain (e.g., the hearing aid's gain) and filtering the audio frequency range in which the feedback occurs; however, these solutions also reduce the effectiveness of the hearing aid or sound enhancement system in amplifying the desired signal. The feedback gain can also be mechanically reduced by fitting an earmold into the user's ear, but many users find this solution uncomfortable and unsightly.

[0016] In a conventional method for feedback cancellation, the Amplitude Slope Deviation (MSD) method is used to detect candidate frequencies. In the MSD method, it is assumed that in the frequency domain, the power of a howling component (e.g., a frequency bin) increases linearly with time when the power is given in decibels. For the identification process, power values ​​(in decibels) are stored sequentially in a finite-length array. A reference gradient is calculated using the first and last values ​​in the array. The gradient values ​​of adjacent amplitude values ​​are then calculated and compared to this reference gradient to find the deviation between the gradient values. If the average deviation between gradients is less than a certain threshold, the frequency bin in question is marked as a possible howling frequency.

[0017] For candidate frequencies below, for example, 4kHz, if the fundamental frequency and its harmonics are absent, the given candidate is likely to produce a howling sound. A notch filter is applied at the given candidate frequency to eliminate the howling.

[0018] The aforementioned lack of harmonics can be part of the Peak-Harmonic Power Ratio (PHPR) method, in which the howling signal is assumed to be of pure tone nature, without significantly higher harmonics. For the identification process, the power ratios between the suspected howling frequency and its harmonics and subharmonics are calculated; these power ratios constitute a single ratio-related value. If this ratio is greater than a certain threshold, the frequency in question is marked as a possible howling frequency. However, conventional feedback cancellation as described above may not prevent howling, for example, due to excessive gain at a given candidate frequency, and / or when the acoustic transfer function between the speaker and microphone changes faster than the response time of the feedback cancellation method.

[0019] Feedback howling occurs because the processing circuitry used to generate the audio output, along with one or more microphones and speakers, inherently defines an audio feedback loop; and this loop has a resonant frequency, for example, due to the time delay of the digital filters in the processing circuitry.

[0020] Embodiments of the present invention address the root cause of the problem in a novel way by identifying multiple frequency components in the drive signal generated to drive the speaker, these frequency components indicating an impending increase in howling signal. The multiple frequency components are spaced apart from each other by a constant frequency difference, which depends on the system's loop delay. Furthermore, in the embodiments, the dominant frequency component among the multiple frequency components indicates the primary howling frequency.

[0021] In some embodiments, the microphone and speaker are mounted on a frame that is mounted on the user's head. In some embodiments described below, the microphone and speaker are mounted on an eyeglass frame. Alternatively, the microphone and speaker may be mounted on other types of frames or head-mounted devices (HMDs), such as virtual reality (VR) or augmented reality (AR) head-mounted devices, or in other types of mounting arrangements. Further alternatively, the microphone and speaker may be used in a sound enhancement system. In this context, a sound enhancement system refers to an audio system that includes one or more microphones, one or more speakers, and signal processing capabilities. The system picks up audio sounds via the microphones, processes these sounds using signal processing methods, and provides the processed signal to the speakers. For example, a sound enhancement system may be used in a room or on a stage for a large audience.

[0022] In this context, an HMD includes any type of frame on which microphones and speakers can be mounted. An HMD can be selected from a list including (but not limited to) eye-wearing devices, eyeglasses, eyeglass frames, goggles, helmets, face shields, headbands, and clip-on devices. In some embodiments, one or more microphones are mounted on the front of the frame, and speakers are mounted on the frame near the subject's ears.

[0023] Consider an audio system comprising one or more microphones, speakers, and processing circuitry. The microphones are configured to output electrical signals in response to sound waves incident on them. The processing circuitry is configured to amplify and filter the electrical signals using digital filters (e.g., beamforming filters incorporating feedback mitigation methods) to generate drive signals for input to the speakers. It detects a growing feedback signal by identifying multiple frequency components in the drive signal that are spaced constantly apart from each other (in some embodiments, this frequency difference is inversely proportional to the loop delay of the system) and adjusts the drive signal to suppress the feedback signal.

[0024] In mitigating howling, the processing circuit is configured to identify the dominant frequency component among multiple frequency components and suppress the howling signal by reducing the gain of at least the dominant frequency component. Furthermore, the processing circuit is configured to gradually increase the gain of the dominant frequency component after reducing the gain, while verifying that the drive signal does not contain multiple frequency components.

[0025] In an embodiment, in order to control the gain of a frequency range containing the howling frequency, the digital filter includes a plurality of sub-filters having corresponding frequency bands and gains, and the processing circuitry is configured to suppress the howling signal by reducing the gain of the sub-filters whose frequency bands contain at least the dominant frequency component.

[0026] In an embodiment, to detect howling signals using a time-domain method, the processing circuitry is configured to identify that the drive signal contains multiple frequency components by: (i) calculating a normalized autocorrelation signal within a specified hysteresis range based on the drive signal; (ii) calculating the maximum value of the normalized autocorrelation signal within the hysteresis range; and (iii) detecting whether the maximum value exceeds a predefined threshold. The processing circuitry can transform the normalized autocorrelation signal to the frequency domain and estimate the howling frequency of the howling signal by identifying the dominant frequency component in the frequency domain. Furthermore, where applicable, the processing circuitry can estimate the loop delay by finding the hysteresis value within the hysteresis range that maximizes the normalized autocorrelation signal.

[0027] In the disclosed technology, increasing howling is detected by identifying multiple frequency components in the speaker's drive signal that have constant frequency differences between them. A dominant frequency component with the howling frequency is detected and suppressed by reducing its gain. In some embodiments, the dominant frequency component can be detected efficiently by applying an autocorrelation technique to the drive signal over a short hysteresis range. This allows for rapid detection of the howling signal before it becomes audible, thus significantly improving the user experience.

[0028] System Description Figure 1 This is a schematic illustration of a hearing aid 20 integrated into an eyeglass frame 22 according to an embodiment described herein. An array of microphones 23, 24 is mounted at corresponding positions on the eyeglass frame 22 and outputs electrical signals in response to sound waves incident on the microphones. In the illustrated example, microphone 23 is mounted on the front part 30 of the frame 22, while microphone 24 is mounted on the temple 32, which is connected to a corresponding edge of the front part 30. Although Figure 1 The large array of microphones 23 and 24 shown is useful in some applications of the invention, but the principles of signal processing and hearing assistance described herein can also be applied alternatively using a smaller number of microphones (with necessary modifications). For example, these principles can be applied using the array of microphones 23 on the front piece 30, as well as in devices using microphone mounting arrangements that are not necessarily based on eyeglasses.

[0029] The processing circuit 26 is fixed within or otherwise connected to the eyeglass frame 22 and coupled via wires 27 (such as traces on a flexible printed circuit) to receive electrical signals output from microphones 23 and 24. Although in Figure 1 The processing circuitry 26 is shown, but for simplicity, some or all of the processing circuitry may alternatively be located in the front element 30 or in a unit externally connected to the frame 22, at some point in the temple 32. The processing circuitry 26 mixes the signal from the microphone to generate an audio output with a specific directional response, for example by applying beamforming to enhance sound originating from a selected angular range while suppressing background sound originating outside that range. Typically, although not strictly necessary, the directional response is aligned with the angular orientation of the frame 22. The processing circuitry also suppresses acoustic signals originating from the speaker picked up by the microphone. In some embodiments, beamforming is omitted from the digital filter, for example, when only one microphone is used.

[0030] The signal processing functions of processing circuit 26 will be described in more detail below.

[0031] Processing circuitry 26 can deliver audio output to the user's ears via any suitable type of interface and speaker. In the illustrated embodiment, the audio output is generated by drive signals for driving one or more audio speakers 28 mounted on the temples 32, typically close to the user's ears. Although in Figure 1 The diagram shows only a single speaker 28 on each temple 32, but the device 20 may alternatively include only a single speaker on one of the temples 32, or the device 20 may include two or more speakers mounted on one or both of the temples 32. In the latter case, the processing circuitry 26 may apply beamforming functionality to the drive signal (using a digital filter that supports beamforming) to direct sound waves from the speaker to the user's ear. Alternatively, the drive signal may be transmitted to a speaker inserted into the ear, or it may be transmitted wirelessly (e.g., as a magnetic signal) to a pickup coil in a user's hearing aid (not shown) wearing eyeglasses.

[0032] Signal processing Figure 2 This is a block diagram schematically illustrating details of the processing circuitry 26 in the hearing aid device 20 according to an embodiment described herein. The processing circuitry 26 may be implemented in a single integrated circuit chip, or alternatively, the functionality of the processing circuitry 26 may be distributed among multiple chips, which may be located inside or outside the eyeglass frame 22. Although in Figure 2 A particular implementation is shown, but the processing circuit 26 may alternatively include any suitable combination of analog and digital hardware circuitry, as well as a suitable interface for receiving electrical signals output by the microphones 23, 24 and outputting drive signals to the speaker 28.

[0033] In this embodiment, microphones 23 and 24 include an integrated analog-to-digital converter that outputs digital audio signals to processing circuitry 26. Alternatively, processing circuitry 26 may include an analog-to-digital converter for converting the analog output of the microphones into digital form. Processing circuitry 26 typically includes suitable programmable logic unit 40, such as a digital signal processor (DSP) or gate array, which implements necessary filtering and mixing functions, as well as feedback cancellation and howling detection and suppression functions, to generate and output digital drive signals for speaker 28.

[0034] These filtering and mixing functions typically involve the application of a digital filter 42 (e.g., a beamforming filter), the coefficients of which are selected to produce a desired directional response. Variations of the digital filter 42 with beamforming functionality are also referred to herein as “beamforming filters.” Specifically, in some embodiments, the coefficients of the beamforming filter 42 are calculated to enhance the sound striking the frame 22 (and therefore the microphones 23, 24) within a selected angular range. Details of filters that can be used for beamforming purposes are further described below. It is worth noting that beamforming functionality can be omitted when the audio system comprises a single microphone.

[0035] Alternatively or additionally, processing circuitry 26 may include a neural network (not shown) trained to determine and apply coefficients to be used in beamforming filter 42. Further alternatively or additionally, processing circuitry 26 may include a microprocessor programmed with software or firmware to perform at least some of the functions described herein. The software may be downloaded to a computer, for example, electronically via a network, or may alternatively or additionally be provided and / or stored on a non-transitory tangible medium, such as magnetic, optical, or electronic memory.

[0036] In implementing beamforming filter 42, processing circuitry 26 can apply any suitable beamforming function known in the art in the time or frequency domain. For example, a minimum variance distortionless response (MVDR) beamforming algorithm can be applied to achieve better spatial resolution. Other applicable beamforming techniques are based on linearly constrained minimum variance (LCMV) and generalized sidelobe canceller (GSC) algorithms. Such beamforming algorithms and other audio enhancement functions that can be applied by processing circuitry 26 are described, for example, in PCT International Publication WO 2017 / 158507 (particularly pages 10-11) and U.S. Patent 10,567,888 (particularly column 9).

[0037] In some embodiments, the processing circuitry 26 includes a feedback canceller 44 that suppresses acoustic feedback from the speaker to the microphone. The goal of the feedback canceller is to prevent howling artifacts by reducing the amount of feedback signal within the signal generated by the microphone. The feedback canceller can be implemented using adaptive filters known in the art. In some cases, the acoustic transfer function between the speaker and the microphone varies over time. For time-varying transfer functions, the feedback canceller may not adapt quickly enough and may produce howling.

[0038] The processing circuitry includes a howling detector 45, which detects and suppresses howling signals that are amplified in the system and are not suppressed by the feedback canceller. The howling detector 45 will be described in more detail below.

[0039] Audio output circuitry 46 (e.g., including a suitable codec and digital-to-analog converter (DAC)) converts the digital drive signal output from beamforming filter 42 (or feedback canceller 44 following beamforming filter) into analog form. Analog filter 48 performs further filtering and analog amplification to optimize the analog drive signal to speaker 28.

[0040] Control circuitry 50 (such as an embedded microcontroller) controls the programmable functions and parameters of processing circuitry 26, which may include a feedback canceller 44. Communication interface 52 (e.g., Bluetooth® or another wireless interface) allows users and / or audiologists to set and adjust these parameters as needed. Power supply circuitry 54 (such as a battery inserted into the temple 32) provides power to other components of the processing circuitry.

[0041] although Figure 2 Primarily referring to hearing devices, the processing circuitry 26 (or a suitable variation thereof) can also be similarly used in other audio systems, such as sound enhancement systems. An example sound enhancement system of this type may include one or more microphones 23, a digital filter 42, a feedback detector 45, a DAC 46, an analog filter 48, and a speaker 28.

[0042] Suppression of feedback howling As described above, the howling in the output of speaker 28 may be due to acoustic feedback from the speaker back to microphones 23 and 24. Howling caused by acoustic feedback is only amplified when the wavelength of the howling frequency is an integer multiple of the length of the acoustic resonator created by filter 42 and other components of processing circuitry 26. The length of the acoustic resonator in the time domain is equal to the feedback delay (also known as the "loop delay" of hearing aid 20), which is equal to the internal delay of processing circuitry 26. ) plus the acoustic propagation time from speaker 28 to one of microphones 23 or 24 ( The following is given: Whistling signals can occur at resonant frequencies The resonant frequency is enhanced, generating a standing wave (with amplitude variation) inside the acoustic resonator: For example, assuming the hearing aid 20 has an internal delay of 8 ms and an acoustic path length of 0.13 ms between the speaker and the microphone, possible feedback frequencies include: Frequency components Constantly spaced frequency difference As shown below: This type of frequency pattern can indicate that the howling signal is increasing because the signal output from the speaker travels an integer number of times along the loop path, causing constructive interference. For example, if a given frequency With a loop gain of 1.1, the amplitude of the interference howl will increase by 10% per round, and may therefore become audible after several loop delay rounds. This leaves enough processing time to detect and suppress the howl signal before it becomes significantly audible.

[0043] Figure 3A and Figure 3B This is a schematic illustration of an example howling signal 60 driving a loudspeaker according to an embodiment described herein, and a corresponding frequency density spectrum 66. In this example, Figure 3A and Figure 3B The signals described herein are generated within a hearing aid such as hearing aid 20.

[0044] In this example, the howling signal has a sampling rate of 16 kHz, and the depicted time period is approximately 30 ms. The depicted example howling signal has a dominant howling component at a frequency of 2.1 kHz, and a loop delay of 8.125 ms. Figure 3A As shown, in the time domain, the howling signal is similar to an amplitude-modulated signal, where the fast internal signal (62) corresponds to the dominant howling frequency (centered at 2.1 kHz), and the frequency of the envelope (64) corresponds to the reciprocal of the average loop delay, which is 123 Hz in this example.

[0045] like Figure 3B As shown, in the frequency domain (66), the howling signal contains multiple frequency components spaced by a constant frequency difference Δf, where Δf = 1 / 8.125ms = 123Hz. In this example, the dominant frequency component among the multiple frequency components has a frequency of 2.1kHz.

[0046] Methods for detecting and suppressing howling signals Figure 4 This is a schematic flowchart illustrating a method for howling detection and suppression according to an embodiment described herein. The method will be described as being performed by processing circuitry 26 of a hearing aid 20.

[0047] The method begins with a drive signal generation step 100, where processing circuit 26 amplifies and filters electrical signals received from microphones 23 and 24 (e.g., using digital filter 42) to generate a drive signal for input to speaker 28 (e.g., via output audio circuit 46 and analog filter 48).

[0048] In signal monitoring step 104, processing circuit 26 monitors the drive signal of step 100 to detect that the howling signal is increasing by identifying multiple frequency components in the drive signal that are spaced apart by a constant frequency difference from each other. Identifying that the drive signal contains multiple frequency components can be done in the frequency domain or the time domain.

[0049] In suppression step 108, in response to the detection of a howling signal in step 104, the processing circuit adjusts the drive signal to suppress the howling signal. For example, if the processing circuit identifies a dominant frequency component among multiple frequency components in step 104, the processing circuit suppresses the howling signal by reducing the gain of at least the dominant frequency component.

[0050] For example, the processing circuitry can suppress the howling signal by reducing the gain of the digital filter 42 within a predefined frequency range centered on the dominant frequency component in the drive signal. Since the width of the howling tone (relative to the half-power point of the howling tone peak) is typically on the order of tens of hertz, a suitable frequency range of tens of hertz covering the width of the howling tone can be used.

[0051] In some embodiments, in step 108, and after reducing the gain, the processing circuitry gradually increases the gain of the dominant frequency component (and possibly other frequency components, if applicable) while verifying that the drive signal does not contain multiple frequency components (e.g., by looping back to step 100 and then to step 104). This mechanism allows for a return to the default gain when the conditions causing the howling are no longer valid.

[0052] At least the gain of the dominant frequency can be controlled, for example, by modifying the coefficients of the digital filter to correspondingly modify the filter's transfer function. In an alternative embodiment, the digital filter comprises multiple sub-filters with corresponding frequency bands and gains. In these embodiments, the processing circuitry suppresses the howling signal by reducing the gain of the sub-filters whose frequency bands contain at least the dominant frequency component.

[0053] After step 108, the processing circuit 26 loops back to step 100 to apply beamforming to subsequent electrical signals received from the microphone.

[0054] Figure 5 The flowchart schematically illustrates a method for detecting and suppressing howling signals based on computational time-domain autocorrelation according to embodiments described herein.

[0055] This method can be performed by elements of the processing circuitry 26 of the hearing aid device 20, and specifically by the feedback detector 45. The method begins at frame reception step 120, where the feedback detector 45 receives a frame of the drive signal that drives the speaker (28) of the hearing aid device 20. This frame can contain any suitable number of “N” digital samples. In this example, the frame contains N=16 samples of the drive signal, which is sampled at a sampling rate of 16 kHz.

[0056] In autocorrelation calculation step 124, the howling detector calculates the normalized autocorrelation signal for the frame within a specified hysteresis range. As shown below: in, This represents the driving signal in the time domain ("n" is the time index). Indicates by This represents the normalized autocorrelation value at lag "m" within the lag range. In this example, the normalized autocorrelation signal... The value is taken in the range [-1, 1], and the maximum value is expected to be found at the hysteresis value corresponding to (or closest to) the loop delay. Since the loop delay is known to be approximately 8 ms in this example, the hysteresis value corresponding to this loop delay is 8 ms · 16 kHz = 128. To find the actual loop delay, the maximum value is searched within the short hysteresis range. A single value might suffice, for example, 16 hysteresis values ​​within the hysteresis range given by 119 ≤ m ≤ 134. When the loop delay is known in advance, the hysteresis range can be determined based on that loop delay. Alternatively, any other suitable hysteresis range can be used.

[0057] In step 128 of the maximum value calculation, the howling detector calculates the maximum value of the normalized autocorrelation within the hysteresis range. This maximum value is also known as the harmonic ratio (HR), and is given below: In detection step 132, the howling detector compares the HR value calculated in step 128 with a predefined threshold represented as HR_thresh. When HR > HR_thresh, the howling detector detects a howling signal and activates howling suppression in activation step 136. Otherwise, the howling detector disables howling suppression in deactivation step 140. After each of steps 136 and 140, the howling detector loops back to step 120 to receive subsequent frames of the drive signal. See below for reference. Figure 6 Describe the methods used to activate / deactivate suppression.

[0058] In some embodiments, the howling detector performs the query of step 132 on multiple frames to achieve a smoother transition between activation and deactivation. For example, the howling detector checks the condition HR>HR_thresh (mentioned above) on a set of multiple recent frames, and activates (or deactivates) suppression if the condition is met in a predefined number of frames in the set.

[0059] Figure 6 This is a flowchart schematically illustrating a method for activating / deactivating howling suppression according to an embodiment described herein.

[0060] This method can be used, for example, to implement Figure 5 Steps 136 and 140 of the method.

[0061] In description Figure 6 In this method, it is assumed that the digital filter 42 includes multiple sub-filters (not shown) with corresponding frequency bands and gains, wherein the gains can be controlled by a howling detector.

[0062] The method begins with the howling detector 45 receiving the normalized autocorrelation signal (in Figure 5 The method calculates step 124 and transforms the autocorrelation signal to the frequency domain. The transformation from the time domain to the frequency domain can be performed, for example, by applying a Discrete Fourier Transform (DFT) to the time-domain autocorrelation signal, such as using a Fast Fourier Transform (FFT) method. Alternatively, other suitable methods for extracting frequency information from the time-domain signal can be used.

[0063] In the howling frequency estimation step 154, the howling detector estimates the howling frequency by selecting the dominant frequency component in the frequency domain.

[0064] In the activation / deactivation suppression step 158, the howling detector decreases (for suppression activation) or increases (for suppression deactivation) the gain of the sub-filter in the band containing the estimated howling frequency. In this embodiment, the howling detector applies the suppression deactivation operation only after suppression activation. After step 158, the method terminates.

[0065] In some embodiments, after activation suppression, the howling detector drastically reduces the gain of the correlation sub-filter to prevent howling signal amplification. In some embodiments, after deactivation suppression, the howling detector gradually increases the gain of the correlation sub-filter, for example, in Figure 5 The method is performed over multiple loops. In an example embodiment, for this purpose, the howling detector may hold a predefined sequence of gain values ​​that increases along the sequence. In an alternative embodiment, the howling detector may select an increasing gain value on an ad-hoc basis for each loop.

[0066] In some embodiments, howling from speech signals with relatively low pitch frequencies may lead to false detections, for example, when using the howling detection method described above. In this case, the howling detector can apply a high-pass filter to the drive signal to suppress low-frequency speech energy and preserve the relatively high howling energy at higher frequencies. In another embodiment, false detections can be avoided by calculating the correlation between frames, since the frequency of howling signals is more fixed in time than the pitch, which typically varies over time.

[0067] Consider the example sequence of operations used to mitigate feedback using the above embodiments: (1) A subject using a hearing device places his hand between the speaker and the microphone, which alters the acoustic transfer function from the speaker to the microphone. (2) As a result, the feedback signal begins to intensify at a certain feedback frequency. (3) The feedback detector detects the feedback before it becomes audible, thus maintaining a good user experience. (4) The feedback detector activates feedback suppression to reduce the amplification gain at least at that feedback frequency. (5) When feedback is not detected, the feedback detector gradually reduces the amplification gain at the feedback frequency to its normal level.

[0068] Example signals related to howling detection Figure 7 The embodiments described herein are illustrated schematically. Figure 3A The diagram shows the howling signal and the time frame used to calculate the normalized autocorrelation signal.

[0069] This illustration depicts frame 170 with a length of N=16 samples, for which the howling detector calculates normalized autocorrelation (e.g., as shown above). Figure 5 (as described in step 124 of the method). In this example, frame 172 of 16 samples is used to calculate the autocorrelation with minimum lag for 119 samples, and frame 174 of 16 samples is used to calculate the autocorrelation with maximum lag for 134 samples. For clarity, other frames for lag values ​​in the range of 119-134 are omitted.

[0070] Figure 8 The embodiments described herein are illustrated schematically. Figure 7 Normalized autocorrelation signal of the howling signal The illustration at 176. In this example, the normalized autocorrelation signal is given for hysteresis values ​​in the range of m = 119…134. As shown in the figure, the normalized autocorrelation signal… It is periodic, with a period of about 8 samples (due to the whistling frequency of 2.1KHz). The signal has a maximum value close to 1 at a hysteresis of m=130, which means that the estimated loop delay is given by 130 / 16KHz=8.125ms.

[0071] Figure 9 This schematically illustrates the transformation to the frequency domain according to the embodiments described herein. Figure 8 A diagram illustrating the normalized autocorrelation signal.

[0072] Frequency domain signal 178 is depicted for a frequency range of 0-8 kHz. As shown, frequency domain signal 178, corresponding to normalized autocorrelation signal 176, has a dominant frequency component at 2 kHz. Therefore, in some embodiments, the howling detector can estimate the howling frequency by finding the dominant frequency component in the frequency domain.

[0073] The above embodiments are given by way of example, and other suitable embodiments may also be used.

[0074] For example, in the above embodiments, howling detection is based on calculating the time-domain autocorrelation signal. However, in alternative embodiments, howling signals can be detected by identifying frequency components in the frequency domain that are spaced by constant frequency differences using any suitable spectral analysis method.

[0075] As another example, the howling detection and suppression functions in the above embodiments can be implemented in conjunction with the above beamforming technology or independently of the above beamforming technology.

[0076] Although the embodiments described herein (e.g., Figure 4 , Figure 5 and Figure 6 The methods described herein primarily address the prevention of feedback in hearing aid devices, but the methods and systems described herein can also be used in other applications, such as any sound enhancement system that may employ large loudspeakers.

[0077] It should be understood that the above embodiments are cited by way of example, and the appended claims are not limited to what has been specifically shown and described above. Rather, the scope includes combinations and sub-combinations of the various features described above, as well as variations and modifications of these features, which would be apparent to a person skilled in the art upon reading the above description and are not disclosed in the prior art. Documents incorporated herein by reference are considered part of this application, and the definitions in this specification should be considered only, unless any terms are defined in these incorporated documents in a manner that conflicts to some extent with the definitions expressly or implicitly made in this specification.

Claims

1. An audio system, comprising: One or more microphones are configured to output electrical signals in response to sound waves incident on the microphones; speaker; and The processing circuit is configured to amplify and filter the electrical signal using a digital filter to generate a drive signal for input to the speaker, detect that the howling signal is increasing by identifying multiple frequency components in the drive signal that are spaced at constant frequency differences from each other, and adjust the drive signal to suppress the howling signal.

2. The audio system according to claim 1, wherein, The frequency difference is inversely proportional to the loop delay of the system.

3. The audio system according to claim 1, wherein, The processing circuit is configured to identify the dominant frequency component among the plurality of frequency components and suppress the howling signal by reducing the gain of at least the dominant frequency component.

4. The audio system according to claim 3, wherein, The processing circuit is configured to gradually increase the gain of the dominant frequency component after decreasing the gain, while verifying that the drive signal does not contain the plurality of frequency components.

5. The audio system according to claim 3, wherein, The digital filter includes multiple sub-filters with corresponding frequency bands and gains, and the processing circuitry is configured to suppress the howling signal by reducing the gain of the sub-filters whose frequency bands contain at least the dominant frequency component.

6. The audio system according to any one of claims 1-5, wherein, The processing circuit is configured to identify that the drive signal contains the plurality of frequency components by: (i) calculating a normalized autocorrelation signal within a specified hysteresis range based on the drive signal, (ii) calculating the maximum value of the normalized autocorrelation signal within the hysteresis range, and (iii) detecting that the maximum value exceeds a predefined threshold.

7. The audio system according to any one of claims 1-5, wherein, The digital filter includes a beamforming filter.

8. A method for audio processing, comprising: Install one or more microphones, which output electrical signals in response to sound waves incident on the microphone; Install speakers; The electrical signal is amplified and filtered using a digital filter to generate a drive signal for input to the loudspeaker; The whistling signal is detected to be increasing by identifying multiple frequency components in the drive signal that are spaced at constant frequency differences from each other; and Adjust the drive signal to suppress the howling signal.

9. The method according to claim 8, wherein, The frequency difference is inversely proportional to the loop delay of the system.

10. The method of claim 8, further comprising: The dominant frequency component among the plurality of frequency components is identified, and the howling signal is suppressed by reducing the gain of at least the dominant frequency component.

11. The method of claim 10, further comprising: After reducing the gain, the gain of the dominant frequency component is gradually increased while verifying that the drive signal does not contain the multiple frequency components.

12. The method according to claim 10, wherein, The digital filter includes multiple sub-filters with corresponding frequency bands and gains, and wherein suppressing the howling signal includes reducing the gain of the sub-filter whose frequency band contains at least the dominant frequency component.

13. The method according to any one of claims 8-12, wherein, Identifying the plurality of frequency components in the drive signal includes: (i) calculating a normalized autocorrelation signal within a specified hysteresis range based on the drive signal, (ii) calculating the maximum value of the normalized autocorrelation signal within the hysteresis range, and (iii) detecting that the maximum value exceeds a predefined threshold.

14. A head-mounted device (HMD), comprising: A frame, which is configured to be mounted on the subject's head; One or more microphones are mounted on the frame and configured to output electrical signals in response to sound waves incident on the microphones; A loudspeaker, which is mounted on the frame; and The processing circuit is configured to amplify and filter the electrical signal using a digital filter to generate a drive signal for input to the speaker, detect that the howling signal is increasing by identifying multiple frequency components in the drive signal that are spaced at constant frequency differences from each other, and adjust the drive signal to suppress the howling signal.

15. The HMD according to claim 14, wherein, The HMD includes devices selected from a list including: eye-wearing devices, eyeglasses, eyeglass frames, goggles, helmets, face shields, headbands, and clip-on devices.

Images