JP2002519973A - Audio processing method and apparatus - Google Patents

Abstract

(57) [Summary] Detect sound at at least two separated detection positions; analyze the detected sound to identify an angular relationship between each sound source and the detection position; Causing the selection of an associated angular relationship; and processing the detected audio in response to the selection to enhance the flow of audio associated with the particular sound source. This method can be used for a hearing aid that minimizes background noise by allowing a user to play a sound by bidirectionally selecting a specific sound source.

Description

DETAILED DESCRIPTION OF THE INVENTION

TECHNICAL FIELD OF THE INVENTION [0001] The present invention relates to a method and apparatus for processing speech, and more particularly, but not exclusively, to a hearing aid, particularly a bidirectional hearing aid.

BACKGROUND OF THE INVENTION With respect to modern hearing aids, (i) using a single microphone to detect speech, and (ii) selectively transforming input speech, and potentially more sophisticated First convert the audio to digital form to take advantage of digital signal processing techniques, and (iii) convert the audio back in the ear canal (or cochlea stimulator).
In the case of r implants), the problem of hearing loss is addressed by directly stimulating the nerves of the spiral ganglia of the organ of Corti. With the use of a single microphone, selectively amplifying sound arriving from a particular direction is achieved only by utilizing a microphone that is directional in sensitivity. However, using microphones with high directivity introduces another problem in that sound coming from a certain range of other directions cannot be detected. In an embodiment of the present invention, two microphones are used and different sound sources (or streams) present in the incident sound.
Embodies a method that is believed to be used in animal hearing systems to identify sounds from the animal.

Even with the use of more than two microphones, existing systems have little ability for a user to interact bidirectionally with the system to select the optimal audio processing. This is a drawback due to the nature of the problem that whether or not the user can judge the voice strongly depends on the situation where the user is placed. The situation changes from a quiet environment with only one sound source to a noisy room with many sound sources. The basic information used by the auditory system to determine the direction of the sound source is the interaural intensity differences (internal intensity differences; IIDs)
) And interaural time differences (internal ear differences: ITDs)
It is. To determine IID or ITD, the hearing aid must have more than one microphone.

[0004] US Pat. Nos. 3,946,168 and 3,975,599 essentially use the different directional characteristics of two microphones, being switched between differently directed and directional input signals. Two in a single container
Hearing aids using two microphones are disclosed. Similar ones utilize both omni-directional and directional microphones in US Pat. No. 5,524,056 and include some suitable equalization means. More sophisticated is the use of multiple microphones that are placed on the user's body (at the frequency of interest) and separated by half a wavelength, as in US Pat. No. 4,751,738. Signals from these microphones are processed by an adder, a bandpass filter, and an amplifier. Thereby, directivity is given to the selected frequency band in the direction in which the user is facing. Further, U.S. Pat. No. 5,737,430 is extended to wirelessly connect a signal from a microphone to a hearing aid placed on the ear.

With the advent of portable digital signal processing (DSP), more sophisticated signal processing methods can be adopted. DSP technology is a binaural system (per ear,
(With two microphones and two output transducers) (U.S. Pat. No. 5,479,522, U.S. Pat. No. 5,651,071) to provide a system that selectively amplifies the signal characteristics of audio while allowing the user to Maintain the exact timing of the signal to be able to detect the direction of the sound source. In this way, the system also achieves noise reduction. Directivity has been added by using DSP technology to perform beamforming (US Pat. No. 5,511,128). U.S. Pat. No. 5,757,932 describes an implementation of these techniques using wireless communication. Binaural II for detecting the direction of the sound source while giving the maximum directivity to the response
Techniques that attempt to resolve the conflicting objectives of preserving D and ITD are compared by Deslogue et al. (JG Deslogue, WM Ravinowitz and PM Perek, binaural microphone array hearing aids-Part 1: Fixed Processing system, IEICE Bulletin on Conversation and Audible Processing, 5 (6): 529--542
Page, published in 1997). Algorithms that attempt to amplify only those sounds using IIDs and ITDs suitable for frontal sound sources are described by Colemeier et al. (B. Colemeier, J. Paissig and V. Hoffman, Frequency Domain). Configuration of a binaural noise reduction hearing aid that performs real-time processing, Scandinavian audiology: Supplement, 38: 28--28, 1993.

[0006] Other multi-microphone technologies use sub-band adaptive processing, pioneered by the University of Paisley, Scotland, which allows for statistically different signals (such as voice and car noise) to be generated. Separation is possible and signal-to-noise ratio (SNR) is improved (PW Shields and D. Campbell, multi-microphone subband adaptive signal processing for improved hearing aid performance: using volunteers with normal hearing) Preliminary test results; ProcICASSP97, pages I, 415-418,
1997; P. Shields, M. Gilorami, D. Campbell, and C. Feife, binaural unmass stripping for emphasis on noise and reverberant speech
Adaptive processing mechanism inspired by king); LS Smith and A. Hamilton, eds., Neuromorphology: Engineering Silicon from Neurobiology, 61--
74, World Scientific, 1998; A. Hussein and D. Campbell, Binaural Subband Adaptive Speech Enhancement Using Human Cochlea Model and Artificial Neural Network, LS Smith and A. Hamilton, Ed., Neuromorphology System : Engineering Silicon from Neurobiology, 75--86, World Scientific, 1998). In addition, blind signal deconvolution
(Unique composition analysis) Anti-heavy learning techniques from training can be used and different sound streams can be recovered.

[0007] Both normal and hearing impaired listeners can move their heads to assist in selecting a sound source of interest. Since the first hearing aid was mechanical, the user had to move the hearing aid to change the direction. With the first electrical hearing aids, little user interface was added, so it was necessary to switch them on and off, and possibly to do multiple settings and even volume control. Modern hearing aids are configurable and are tuned to compensate for hearing loss specific to each user, but typically cannot be readjusted thereafter by the user.

[0008] Some recent hearing aids have added the ability to change settings. US Patent 5,
The particular microphone used in the invention described in 524,056 can vary. U.S. Pat. No. 5,636,285 similarly describes a technique for re-configurable voice control.

[0009] Most hearing aids retransmit sound into the ear canal. Measures to properly select and convert between incoming and outgoing speech are based on replenishing the hearing range that the user considers to have lost. However, partial hearing loss not only reduces sensitivity in some frequency ranges. If it were just such, this approach would have been entirely successful. The problem is that most of the attenuation in sensitivity is due to a defect in the conversion system of the auditory cells, especially in the basal (high frequency) part of the cochlea. Simply amplifying the high frequency signal does not stimulate the auditory nerves distributed inside the auditory cells. Instead, other (undamaged) auditory cells from elsewhere in the cochlea react and mix their own response with the response to the amplified high frequency sound. In this way, selective amplification increases the response of auditory cells to a broader frequency range, but at the expense of its position-specific frequency resolution.

[0010] Furthermore, for patients with many defects, the distance between audible and painful sounds is small and means of adjusting a frequency sensitive amplifier are difficult.

For patients with little or no remnant hearing loss, the technique of cochlear implant (cochlear implant) is often used instead of retransmitting the auditory signal. They directly stimulate the nerve cells of the spiral ganglion. Unfortunately, it is not possible with this method to stimulate all of the auditory nerves because the shape of the cochlea inhibits it.
Thus, only the high frequency (basal) portion of the cochlea is stimulated, giving the user a very weak signal.

Another possibility with little or no remnant hearing (especially when the auditory nerve or brain stem is damaged) is the use of different treatment devices, such as visual treatment devices. Can be This is proposed in U.S. Pat. No. 5,029,216, which describes a system mounted on glasses that can alert a driver with hearing loss to the approach of an ambulance vehicle.

[0013] Whether retransmitting speech, directly stimulating the auditory nerve, or using a visual area, both speech and noise of interest can be introduced. Focusing on those parts of the spectrum that are most likely to occur in conversation, it will amplify the conversations of all speakers at once, and even if the hearing aid user can understand one speaker speaking, This leads to a common problem that cannot be understood when there are speakers. Separating the speech and noise of interest is difficult because the situation may change over time and the listener may move. As a result, both signals and noise of interest to the user tend to be amplified. The multi
The problem that the microphone technology is trying to solve was such a problem. However, its directivity is often limited to the direction in front of the user.

The physiology of the early stages of the auditory system is well known and is well described, for example, in JO Pickles, Introduction to Auditory Physiology, Academic Press, 2nd Edition, 1988. This physiology is very similar across a wide range of mammals, which means that everything that happens at this stage is (i) effective and (ii) that auditory processing is particularly human-specific It does not need to be assumed. What we should pursue is that sound is flowing in one ear or both ears (AS Bregman, Auditory Scene Analysis, MIT Press, 1990).
Suggests. This is because (i) the same streaming problem has been found throughout the animal kingdom, and (ii) logically, audio streaming precedes its interpretation.

The auditory system appears to use a number of different cues in performing the streaming. In binaural streaming, relative intensity, relative timing of the sharp increase in intensity in the frequency domain (onset (rising)), relative timing of the band-passed audio envelope features (most prominent) Amplitude modulation peaks) and the relative timing of the peaks and troughs of the bandpassed signal (synchronous features of the waveform). Relative intensity is most used at low frequencies where head shadows are present in sound from the stronger side of one ear than the other. At high frequencies it is less pronounced because the sound waves diffract around the head. Streaming with one ear appears to use co-occurrence and relative timing over the onset (rising) frequency range, and co-occurrence of the same frequency amplitude modulation in the middle and high frequency regions of the frequency range. This list is not intended to be exhaustive, but provides examples that point in the direction of techniques used simultaneously by early auditory systems.

Apart from the relative strength, all of the features described above are based on the fine temporal structure of the speech. These features are divided into three classes (S. Rosen, Temporal Information in Speech: Acoustic, Auditory and Linguistic Aspects, Phil. Trans. R. Soc. London B, 33
6: 367-373, 1992; LS Smith, features extracted from short-term structure of cochlear filtered sound, JA Ballina Ria, DW Grasspool, H. Hofton, Ed., 4th Neurocomputation and Psychology Seminar London, April 9-11, 1997, 113--125, Springer Varag, 1998), and features found in all three classes are used for grouping and direction finding in one ear Including information that would be

The source of these characteristic differences between the two auditory systems is the time difference between the ears (IT
D) and the difference in sound intensity between both ears (IID). The exact shapes they take are described in J. Brauart, Spatial Hearing, MIT Press, Revised Edition, 1996. It is clear from this discussion that IID is effective at low frequencies (due to head shadow effects) and ITD is effective at mid and high frequencies (the signal spacing is the signal path to each of the two ears). Because it is bigger than the time difference) (
(See FIG. 1).

The auditory cells inside the organ of Corti transform the pressure waves at the basilar membrane of the cochlea, thus spiking the auditory cells at a frequency of at least between about 20 Hz and 4 kHz, at a fraction of the phase of the input signal. Easy to wake up. This phase locking is considered to be important in detecting the difference in the time difference between both ears in the processing of the auditory signal later. As long as the period of the signal is long compared to the ITD, this provides an important and obvious cue.
However, for example, at 2.5 kHz, the period is 400 μsec. ITD (
If 150 μsec (corresponding to θ = 25 degrees), the signals reaching the two ears are:
Expect a 3π / 4 phase shift. However, this is equivalent to 250 at θ = 37 degrees.
It is not identified as a signal having a phase shift of 2π-3π / 4 corresponding to the ITD of μsec. Therefore, the direction becomes ambiguous from the middle to the high frequency.

At high frequencies (approximately 4 kHz and above), phase-locking is disrupted, so waveform-synchronized ITDs are not used to locate certain high-frequency sounds.

[0020] A sudden increase in voice intensity causes firing of the nuclei of the cochlea (causing the effect of stimulus transmission).
(JO Pickles, Introduction to Auditory Physiology, Academic Press, 2nd ed., 1988). These rising cells have a short reaction time (the time from the application of a stimulus to a response) and are relatively insensitive to intensity (JS Rotman, ED Young and PD Manis, ventral cochlea) Convergence of auditory cell fibers on dense cells of the nucleus, inference by computational models, Journal of Neurophysiology, 70 (6): 2562--2583, 1993; JS Rotman and ED Young, ventral cochlear nucleus Increased neuronal entrainment in a computational model of dense cells, Auditory Neuroscience, 2: 47-62, 1996). This is important because the intensities are considered different for the two detectors. Reaction times are very short, and population coding of multiple cells is thought to be able to accurately measure rise timing (DC Fitzpatrick, R. Buttra, TR Stanford and S. Kuuda, Coding of a neural group for confirmation, Nature, 388: 871-
874, 1997; BC Cotton, Voice Localization and Neurons, Nature, 393: 531, 1998). As a result, the rising envelope ITD can provide useful directional information even for high frequency signals.

In a normal environment, most sounds reach the ear via various paths due to the presence of a reflective surface. One consequence of this is that ITD may be the result of a combination of these multiple paths, which results in an inaccurate estimation of the direction of the sound source. However, the straight path is always the fastest and generally has little attenuation. In this way, the direction of the speech calculated at the first rise is not affected by the existence of a plurality of paths. In addition, the rise caused by the arrival of the signal from the reflection path attempts to generate a response from the rising cell that has just fired by the signal from the direct path. These reflected rises are not as strong as the original rise and attempt to stimulate cells that are considered to be in a refractory period.

Many toned sounds consist of multiple harmonics of the low frequency fundamental. This is the reality of the sounds emitted by many animals, including the sounds produced in conversation. as a result,
It is particularly important to be able to detect these directions. The essence of bandpass filtering that occurs in the cochlea is that in these sounds, multiple adjacent harmonics appear at many high frequency locations in the cochlea. This becomes the kinetic energy of the basilar membrane of the cochlea when modulated in amplitude at the frequency of the fundamental. As a result, the auditory nerve output is similarly modulated. Radial (toothed) cells in the nucleus of the cochlea are particularly sensitive to amplitude-modulated signals and can amplify this modulation (AR palmers and IM winters, cochlear nerves and the center of the cochlea, generated in conversations). Response to fundamental frequency and harmonic complex tones of simulated speech, advances in biochemistry, 83:23
1-239-239, 1992). By using the difference in peak and trough timing of this amplitude modulation, the auditory system can detect the direction of constant high frequency audio, even without locking of the waveform synchronization phase.

Recent studies in auditory psychophysics have proposed a model in which ITDs detected from rise, amplitude modulation and waveform synchronization are not directly used but grouped first in one ear. (S. Carlisle, Physics and Psychophysical Fundamentals in Speech Position Recognition in Temporary Auditory Space: Generation and Adaptation
RG Lands Company, 1996, JF Cling and Q. Summerfield, Simultaneous Perceptual Separation of Speech: Lack of Cross-Frequency Grouping by Common Binaural Delay, J. Acoustical Soc. Of America, 98, Vol. 785-796, 1995; CJ Darwin and V. Thiocca, Grouping of Pitch Perception: Ear Asynchrony Effects and Inaccurate Coordinating Ears, J. Acoustical Soc. Of America, 91,
6, 3381-3390, 1992; CJ Darwin and RW Fukin, Perceptual separation of harmonics from vowels by binaural time difference and frequency proximity, J. Accoustical Soc. Of Ameri
ca, 102 (4), 2316-2324, 1997). This seems to be most effective since ITDs are calculated from multiple channels simultaneously, reducing the likelihood of errors.

Real-world ambient sounds are often complex and contain energy at many different frequencies. Some are toned and some are not. However, determining the direction of the different types of sound is usually not felt particularly difficult. The auditory system uses IID (and possibly other unknowns) for all of the above technologies
It is thought that it is because it is used in addition. Indeed, (i) IID is particularly useful for low frequency speech, and (ii) rising is useful for speech that is toned or untoned (such as applause) if the speech occurs rapidly. (Iii) waveform synchronization techniques are useful for intermediate frequency speech, and (iv) amplitude modulation based techniques are useful for high frequency speech that exhibits amplitude modulation when bandpass filtered. is there. (You will find it difficult to find the exact direction of pure constant high frequency speech. The IID is small and waveform synchronization is not possible.) Hearing loss is a problem in all aspects of the fine temporal structure. Causes information loss. Obviously, there is no fine time information in the frequency band of the signal that is not detected at all by the internal auditory cells. Furthermore, in the frequency bands of the signal in the area where the organ of Corti does not function, the detection will only occur in the area adjacent to the organ of Corti. Loss of synchronization of the auditory nerve signal will occur for the characteristics of the envelope modulation and for the peaks and troughs of the bandpassed signal itself. This lack of minute time information is considered to be one of the root causes for a hearing-impaired person who has difficulty streaming sound.

It is an object of an embodiment of the present invention to provide an interactive system in which a classification of audio features is comprehensively detected and used to find the direction of the input audio. Also,
This direction information can be used to stream audio.

[0026] SUMMARY OF THE INVENTION Accordingly, in a preferred embodiment of the present invention to provide a two-way system, the system comprising precise timing of the signal produced by the bandpass input sound at the two microphones, early Are detected using techniques based on what is likely to occur in the auditory system of the subject. Along with the IID on each channel,
Used to determine the direction of the sound source. By interactively selecting several elements to be displayed, the user selects which sound element is to be presented. The user interaction may consist of the user pointing his hand in a particular direction, or may use some form of display and graphic display board. The final representation of the selected auditory information uses the auditory routine (using selective amplification, attenuation and possible resynthesis) or the visual routine.

The system described herein can be used as part of a hearing system for a robot. Sound source from a specific direction, all sound fields (from many sound sources simultaneously)
Rather than having to be interpreted at once, the post-processing of the input sound field
It can be chosen to make it easier.

In one aspect of the invention, a method for processing audio is provided. The method comprises the steps of: detecting sound at at least two isolated detection positions; analyzing the detected sound to identify an angular relationship between an individual sound source and said detection position; Receiving a selection of an angular relationship associated with a particular sound source; processing the detected sound in response to the selection so as to enhance a sound flow associated with the particular sound source.

Preferably, the angular relationship between the respective sound sources is determined at least in part with respect to the time difference between the sounds from the respective sound sources, as detected at the isolated detection position. Most desirably, an angular relationship between the respective sound sources is determined with respect to a detected time difference with respect to at least one characteristic of the detected sound. The features are:
Selected from: pitch phase; amplitude modulation phase;
Ideally, the time difference is detected for a plurality of features of the detected sound.

[0030] Preferably, the angular relationship between the respective sound sources is at least partially detected with respect to a difference in intensity between sounds from the respective sound sources detected at the isolated detection position.

Also preferably, the sound is detected at a position corresponding to the user's ear, and the angular relationship between the respective sound sources is determined with respect to binaural time differences (ITDs) and binaural intensity differences (IIDs).

Also preferably, the method further comprises selectively filtering the sound detected from the isolated location into a plurality of channels, and characterizing the sound of each channel obtained from one location, Includes comparing sound features from corresponding channels associated with other locations.

According to another aspect of the present invention, there is provided a method for processing sound emitted from a plurality of sound sources. The method comprises the steps of: detecting sound at at least two isolated detection positions; from each of the sound sources detected at the isolated detection positions to identify an angular relationship between the individual sound sources and the detection positions. Analyzing the detected sound with reference to at least one of an intensity difference between the sounds of the sound sources and a time difference between the sounds from the respective sound sources as detected at the isolated detection position; and Playing a sound associated with at least one sound source based on the determined angular relationship.

As a further feature of the present invention, there is provided an apparatus for processing sound. The device comprises: means for detecting sound at at least two isolated detection positions; means for analyzing the detected sound to identify an angular relationship between each sound source and said detection position; a specific sound source Means for permitting the selection of an identified angular relationship associated with: and responsive to said selection to enhance the sound flow associated with said particular sound source;
Means for processing the detected sound.

In accordance with yet a further aspect of the present invention, there is provided an apparatus for processing sound emanating from a plurality of sound sources. The device comprises:

Means for detecting sound at at least two isolated detection locations; intensity differences between sounds from each of the sound sources detected at the isolated detection locations; and each detected at the isolated detection locations. With respect to at least one of a time difference between sounds from the sound sources, to determine an angular relationship between the respective sound source and the detection position,
Means for analyzing the detected sound; and means for emitting a sound associated with at least one sound source based on the determined angular relationship.

[0037] These and other features of the invention will become more apparent from the following description when taken in conjunction with the accompanying drawings. FIG. 1 is a graph of the binaural time difference (ITD) as a function of angle from the sound source.
FIG. 2 is a schematic diagram in the form of a block diagram of an apparatus for processing sound according to a preferred embodiment of the present invention; FIG. 2a is a more detailed representation of the output from the bandpass filter of FIG. FIG. 3 is a block diagram illustrating the determination of binaural intensity difference (IID) in the apparatus of FIG. 4, 5 and 6 are block diagrams representing the determination of binaural time differences (ITDs) starting with the device of FIG. 7 to 11 are block diagrams illustrating the determination of ITDs based on amplitude modulated (AM) signals in the apparatus of FIG. FIG. 12 is a block diagram expressing a simple determination of ITDs in the apparatus of FIG.
FIG. 13 is a block diagram expressing the display of IID and ITDs in the device of FIG. 14 to 16 are block diagrams expressing the processing of the user's interaction in the display of FIG.

Referring first to FIG. 1, a binaural time difference (ITD) is defined as the 150 mm inner ear separation (ITD).
Signal path difference) and is graphed as a function of the angle between the source and the eye. It is assumed that the sound source distance is large compared to the distance between the ears. As described in detail, the device of the preferred embodiment of the present invention determines the ITD by a plurality of different paths, and the information is used by the device so that sound can flow toward the user.

FIG. 2 shows an apparatus 10 for processing sound according to a preferred embodiment of the present invention.
FIG. The figure shows two input transducers 12,14 (microphone L and microphone R) and two multi-channel bandpass filters 16,18. The microphones 12, 14 are located in the ear, behind the ear, or in any other suitable location, at a suitable distance. The microphones 12, 14 are of the omni-directional type, and the directivity of the system cannot be achieved with the sensitivity sensitivity of the microphones. Further, although not critical, the microphones are tuned.

Each of the band-pass filters 16 and 18 is connected to each of the microphones 12 and
The electrical signals arriving from 14 are separated into a plurality of bands as shown in FIG. 2a. These bands may overlap and have a wide tuning range. : That is,
They have properties similar to the bands found in sensitivity analysis of the cochlea of real animals. Approximately, at 6 dB, it has a bandwidth of about 10% of the center frequency.

The bandpass filters 16, 18 are matched to each other. In addition, the filters 16, 18 have a fixed and known delay characteristic, which is the same (or very similar) for the two filters 16, 18.

Both bandpass filters 16, 18 have the same number of outputs. The exact number is not critical, but the performance of the system improves as the number of filters increases.

From the bandpass filters 16, 18, the signal characteristics of the individual channels are processed to provide information about binaural intensity differences (IIDs) and binaural time differences (ITDs). The resulting information is provided in a form that allows the user to recognize and select the sound source based on the direction of the sound reaching the user. The signal from the channel essentially associated with the selected sound is processed to suit the special needs of the user, effectively playing the sound from the selected sound source, and the sound or sound from other sound sources. Minimize the effect of "noise".

The operation of the device 10 will be initially described with reference to FIG. 2, followed by a more detailed description in the way that individual features of the sound are detected, analyzed and processed.

In the illustrated embodiment, the outputs from the filters 16, 18 are processed in four different analyses, and the required hardware is illustrated in block form. Both forms of analysis are described shortly and in turn below.

The sound intensity of either channel is calculated at 20, 22 and the determined intensities from the two microphones 12, 14 are compared at 24 on a channel-by-channel basis,
In each channel, a binaural intensity (IID) measurement value is output. Which IID
Also indicate a particular angle between the microphones 12, 14 and the sound source, this information
26 and relayed to an interactive display device 28. As will be described, this display receives similar input from the results of other forms of analysis,
Gives more complete information for the user. The user can act on the display device to select a particular “angle”, and can further select to obtain sound from a sound source corresponding to the angle. The output from the display 28 is relayed to the respective store 26, from which the "contribution" or display details of the channel to the selected angle are extracted and relayed to the resynthesis substation 30. The channel-related information received from storage 26 by substation 3 is used to select and process signals directly from filters 16 and 18, which signals enhance the appropriate channel inputs, and It is selectively processed to provide a signal to the user in a suitable manner, for example, by selectively amplifying selected channels.

In addition to IID, the sensitivity and accuracy of the device is improved by detecting and processing binaural time differences (ITDs) for each channel in three different ways, for the three different methods: This is briefly described below. First, a rise is detected and calculated at each of the microphones 32, 33, and the ITD is at 34,
Each channel is calculated and gated. The resulting rising ITDs are stored at 36 and relayed to the interactive display 28 and resynthesis substation 30 in a manner similar to the IID procedure described above.

Second, the amplitude modulation (AM) in each channel is 38, 39,
The detected information is grouped and the resulting information is used at 40 to calculate and gate AMITD. Again, the resulting AMITDs are stored at 42 and relayed to the interactive display 28 and resynthesis substation 30 in a manner similar to the IID and rising ITD described above.

Third, the signal phase for each channel is detected and grouped at 44 and 46 and the resulting information is used at 48 to calculate the signal phase ITD. Again, the resulting signal phases ITDs are stored at 50 and stored in the interactive display 28 and recombining substation 30 in a manner similar to the IID, rising ITD and AMITD described above.

These operations will be described in more detail first with reference to FIG. In the figure, the calculation of the estimation of the angles is shown, where the dominant input sound is generated by using the binaural intensity difference between the left and right inputs, one estimation number being made per channel. Can be

The IID is calculated at 25 for each channel repeatedly (eg, every 25 ms). The calculated IID is converted at 24 to an estimated number of angles of incidence of the sound using the estimated number of head-related conversion functions. This function is itself a complex function of the frequency of the sound. The estimated angles for each channel are grouped together at 27 to produce multiple estimates of the angle of incidence of the sound. These are then sent to the display subsystem 28.

FIG. 4 shows the rising detector 32 and the rising group detector 32a. There is one rising detector for each band generated by the bandpass filter of FIG. Those from the left edge are processed separately from those from the right edge.

Each of the rise detectors 32 and 33 detects a rise (a sudden increase in energy) in a single channel. The output of the rising detector is a pulse in this embodiment. This pulse is generated very quickly after the onset of energy build-up, and the delay before the pulse is generated does not contribute to the magnitude of the rise. In other words, the potential of the pulse is low. The output of the rising detector is the rising (x, i)
Where x is L or R (left or right) and i specifies a bandpass channel.

The output of the rising detector 32 on one side is supplied to a rising group detector 32a. There is one rising group detector for each microphone (ie, on each side). The riser group detector 32a groups together these risers that occurred in a short time (taking into account the difference in lag times through the filter bank).

The output of the rising group detector 32a is an n-channel signal (n is the number of bandpass bands). Each signal is always 1 (indicating that this channel is now part of a rising group) or 0 (indicating that this channel is not currently part of a rising group). is there.

FIG. 5 shows the left and right rising group signals configured to form a combined rising group signal at 34 a. This synthesis 34a takes the form of a series of n-and-gates or a series of or-gates.

FIG. 6 shows the rising signals from the left and right from each channel compared in time, and calculates the value of the time difference. There are n for which such values are calculated.
No output is generated on channels without a rising signal. Similarly, no output is generated on a channel when the minimum value of the time difference between the left and right rising signals is too large (ie, not generated from any signal direction).

The generated value is 34 b, and is gated by the combined rising signal group signal. This signal selects the rising group (generally one at a time). n outputs are generated, each of which is a time difference value or worthless.

One rising group is generated at a time, and the rising storage unit 36 stores the latest grouped rising channel sets, and these are stored in the grouped ITD.
, That is, indexed in accordance with the direction of the sound determined for each channel.

FIGS. 7 and 8 show how amplitude modulation is detected. The outputs from the bandpass filters 16, 18 are adjusted and smoothed at 60, and the adjusted smoothed output is provided to an AM mapping network 62. This technique is described in Smith L. et al. S. (
One-way frequency map implemented using a network of integrated firing neurons, ICANN 98, Vol. 2, pp. 991-995, Springer Varag, 19
1998). This network 62 (as shown in FIG. 8) has a plurality of stimulatory neurons (m) 64 and inhibitory neurons 66 (hidden). The input of all stimulatory neurons is the same, and the input of inhibitory neurons is the (lagged) output of stimulatory neurons. Stimulatory neurons are arranged to be particularly sensitive to amplitude modulation over a narrow frequency range. The effect of the network is
Due to the amplitude modulation input, one of the stimulating neurons (mapping neurons) fires in phase with amplitude modulation. In addition, inhibitory neurons pulse whenever there is a sufficient amount of amplitude modulation.

The AM selection stage gets the output from all the stimulating neurons. It is gated by the output from the inhibitory neuron (the output is not generated due to the lack of an amplitude modulation input). It reduces the pulse output to a single amplitude modulation channel by selecting only the active stimulatory neuron output. In addition, it encodes the identity of the stimulating neurons that produce this output. This provides information about the frequency of the amplitude modulation.

FIG. 9 shows the generation of a table 68 used to group amplitude modulated signals.
To group the amplitude modulated signals (since they can be used to calculate the grouped ITDs), we use the AM frequency output from the bandpassed channel. 1 for each output for each
Or a table that otherwise contains 0. In the table, each column may have at most one entry. The same AM frequency is found in more than one channel, and thus there are more than one column of one entry. The figure shows the situation for 15 bandpassed channels and 12 AM identifiable AM frequency bands.
In the table shown, bandpassed bands 2, 3 and 9 find AM on AM channel 3 and bandpassed bands 6, 7, 8 and 11 find AM on AM channel 7 and bandpassed. Bands 10, 12 and 14 found AM on AM channel 11.

One table is generated on both sides (left and right), and an integrated table is generated by taking the sum of the left and right tables.

FIG. 10 shows how the columns in the table are used to gate the AM signal at 40, selecting only those with the same AM frequency and binaural time difference (ITDs) generation for comparison. Indicates what is done. In the figure, we show a system with 15 bandpassed channels. The output of column 7 of the above table is used to gate these signals. And only bands 6, 7, 8 and 11 were selected.

These pulse signals have the same frequency (in other words, frequency band 7), and their pulse times reflect the phase of the amplitude modulation, and the pulses are in phase with the amplitude modulation.
, Pairs (left, right) are sent to a circuit 70 which calculates these signal time differences. The selected values across the different bands are processed at 72 (eg, averaged or moded, intermediate, or selected) to convert the AM time difference signal to the AM frequency band of this AM frequency band. Generate for The AM time difference signal is generated in FIG. 9 for each non-zero column, ie, for each of the AM frequency bands detected on both the left and right channels.

FIG. 11 shows how the latest time difference signal generated for each non-zero column is stored. That is, a series of channels associated with each signal are stored and (grouped), such as input from the interactive display 28.
Indexed by ITD.

FIG. 12 shows how simple (ungrouped) ITDs are calculated from the output of each pair (left, right) bandpassed channel. This is achieved by using phase-locked pulse generators 74, 75 (eg, capable of generating zero-crossing pulses toward each positive) and calculating the time difference between these pulses. At low frequencies, these estimated numbers tend to be unreliable, and at high frequencies, they can be ambiguous. However, there are intermediate frequency ranges from which good estimates can be obtained.

One time difference estimation number is generated for each (intermediate frequency) channel. These can be grouped together in preference to further use.

The most recent time difference estimate number is stored at 50, ie, the channels associated with each grouped time difference estimate are stored and indexed by the time difference (ITD) itself.

In FIG. 13, time difference signals from three sound sources (rise (FIG. 6), amplitude modulation (FIG. 11), and waveform synchronization processing (FIG. 12)) are displayed on the display device 28 in the direction display format. That is, the time difference between the left and right channels is interpreted as an angle. Further, the angle calculated from the IIDs (FIG. 3) is displayed.

In this embodiment, the display device has a semicircular shape, and the dark region matches the estimated direction of the sound source, because the system cannot distinguish between the sound before and after. The user communicates on the display and selects a particular direction in which to accept sound (eg, by touching the display). The display returns the selected angle,
This is then processed. Low power flat touch panel displays (such as those used in color portable computers) are available.

FIG. 14 shows how a signal for controlling a signal supplied to the user is generated from information collected from the interactive display device 28, and the information stored in the rising storage unit 36 ( FIG. 6) shows the AM difference signal storage unit 42 (FIG. 11) and the time difference signal storage unit 50 (FIG. 12) based on the stored waveform.

The angle output from the interactive display 28 can be calculated from the user interaction at the display 76.
Is calculated by This translates into an estimated number of IIDs at 78 and 80, to which sound from that direction is directed. The channel contributions from the low, intermediate and high frequencies are normalized at 82, 84 and obtained as a mixed signal.

FIG. 15 shows that the angle calculated from the user interaction in the display is an index into the storage 26 of the contribution of the low frequency channel, and which low frequency (LF) channel is It shows how it is used to estimate what would result in IIDs that would have been generated by a signal from that direction. This uses a transform function associated with a different head at different frequencies.

FIG. 16 shows how the final signal for display to the user is generated.
The mixing signal, control LMix and control RMix, are generated and control the left and right channel mixing units 86, 88 as described in FIG. The final mixing of the signals for the two ears is controlled by the control LMix and control RMix signals in mixing sections 90, 92, which depend on the nature of the user's hearing loss.

As noted above, the present invention seeks to simulate the processing of sound in a human (or other mammalian) auditory system to some extent, as discussed below. The inputs of the system are two microphones 12, 14 (in the figure, L
Is on the left and R is on the right). The microphone is located, for example, at the edge of the middle ear,
Or it is located on the ear wing. Placing the microphone at the edge of the middle ear allows the ear wings to transmit characteristics to change the relative strength of the different frequencies. If the final expression is in both ears, this information is useful to the user,
It allows for better spatial placement of sounds (J. Blauart, Spatial Listening, MIT Press, Revised Edition, 1996). The microphones 12, 14 are
Convert acoustic signals to electrical signals. These electrical signals are amplified (maintaining the same frequency / phase response on both channels) and fed to individual bandpass filter banks 16,18. These filter banks 16, 18 perform the same task as the cochlea. Each of these filter banks 16, 18
Generates many outputs, one for each channel. These channel outputs are
Used as input to a module that mimics a rise detector, a waveform synchronization detector and an amplitude modulation detector of a neurophysiology system. However, not all channels use all three modules.

The earliest studies (LS Smith, segmentation of sounds based on rising, DS Touretzky, MC Moser and ME Hasermo compilation, Neural Information Processing System 8
, Pp. 729-735, MIT Press, 1996) used sound from different channels for sound segmentation. However, the integrated firing neuron model used there has a latent period (time from a sudden increase in neuron firing),
Unlike those of actual rising cell strength, the latency depends on volume and rate of increase (JS Rotman, ED Young and PD Manis, dense cells of the anterior cochlear nerve) Of auditory nerve fibers on the table, predicted effects of computer models, Journal of Neurophysiology, 70 (6): 2562-2583, 19
Published in 1993; S. Rotman and E.L. D. Young, Enhancing Neural Entrainment in a Computer Model of Whole Cochlear Nerve Dense Cells, Auditory Neuroscience, 2: 47-6
2, p. 1996). An overall rise detector is very short, but must have a certain latency. This latency needs to be constant over a wide range of both intensity and rate of increase. Since rising can be used for both toned and untuned sound locations, each rising detector can receive input from a range of bandpassed channels. Unlike biological systems, we use one accurate rising detector per channel, rather than relying on total number coding.

[0078] Waveform synchronization is basically used from low to intermediate frequencies, as discussed earlier. An integrated waveform synchronization detector provides an output at a particular location in the phase of the signal (eg, can generate a zero-crossing pulse towards each positive). Accurate ITD measurements should be as unstable as possible.

[0079] Amplitude modulation is basically useful at intermediate to high frequencies. It should be noted that the effective use of AM is expected for bandpass filters with a wide band response, such as the actual cochlear response. Also, the detector outputs at certain points in the envelope, for example, at peaks, and instability is minimized.

The display device may provide different azimuth directions of the different input sounds (whether the sounds are in front of or behind the user, as calculated from the IIDs and conversion functions associated with the head, and from the ITDs). ). This is itself used to draw attention to features of the auditory environment. However, selecting the information displayed by the hearing aid itself is said to be more useful by allowing the hearing impaired to interact. In a particular application, the best achieved method depends on factors that change from user to user, depending on whether they want to use their hands to interact with the system, or on their heads. It depends on whether you prefer to interact only by turning.

Two main modes of sound selection can be used. In an embodiment preferred by most users, the user turns their face to a sound source of interest (known). The selected sounds are low ITD and IID sounds. In other embodiments, as described above, a map of the input sound is generated and displayed, and the user selects the sound to be displayed.

The information displayed to the user is expressed in one ear or both ears. The following discussion illustrated in FIGS. 11-16 refers to binaural representations. The result of the interaction on the user's interactive display is that if the user wants only these sounds before their eyes,
Is the angle θ between -π / 2 and + π / 2 or 0. This angle is used to calculate the expected IID and ITD for the signal from that direction. In the illustrated embodiment, the ITD estimate is used to index into a stored ITD / channel list for the rise, amplitude modulation and waveform synchronization subsystems. For each channel, these three values are used to calculate an estimate of the channel contribution generated by the sound source from that direction. For each ear, these estimates are normalized (to a length of 1), and the vector (control LMix for the left ear and control RMix for the right ear) controls the mixer (FIG. 16). Used to This produces two composite channel outputs, output data L and output data R. These are used for medium and high frequencies. For low frequencies, the same approach is taken with IID. The resulting output, output data, is a composite channel signal, suitable for visual display. For an auditory display, the signals of the different channels are added together in a manner that reflects the hearing impairment of the user.

Whether the selected sound is correctly displayed to the user depends very much on the user's perceptual ability. If there is sufficient residual hearing, selective amplification is most appropriate. If the residual hearing is limited to a particular frequency band, resynthesis is most appropriate. It is also possible to mix these two techniques. Alternatively, the display can use a visual sensation.

The sounds generated and selected as shown in FIGS. 14 to 16 have a number of channels that are selectively amplified to compensate for hearing impairment. The resulting sound is represented by (a) one ear if sufficient residual hearing is in one ear, or (b) bilateral if sufficient residual hearing is in both ears. You. In this case, the present inventors express the output data L to the left ear and the output data R to the right ear.

In addition, the present inventors have developed a control LRM in which the gains of the signals from the two ears are changed.
ix signal can be used.

When the residual hearing is limited to a small part of the auditory spectrum, or indeed,
When the expression is made through an implant, it is more appropriate to resynthesize the sound in order to obtain the audible benefit. The signal serving as a starting point of the present invention is an output data signal.

If there is little or no residual hearing, the information from the sound will, in one particular direction,
In order to be visually represented and to represent information, color representation is used in such a way that the power of sound is distributed over the spectrum.

One possibility (without using an interactive display, but displaying all incoming sounds) is to choose a color that matches the ITD and create an intensity that reflects the signal intensity.

Alternatively, a bidirectional display can be used to select the direction of the sound source to be represented, the color of the display can be used to indicate the presence and pitch of amplitude modulation, and maintain white in the non-amplitude modulation range of the spectrum. , And can also be used to indicate signal strength.
This uses the information that appears in FIGS. 8 to 10, but is not used for auditory representation.

The two-way system operates under strict real-time forcing. In addition, the system is ideally light, wearable and runs with low power consumption. The most current sophisticated hearing aids are
Uses digital signal processing (DSP) technology. DSP circuits generally function with high-speed parallel processing multiplication and addition. This is highly suitable for complex calculations and highly effective for digital filtering. Non-linear operation is also possible with such a circuit. However, it is desired that DSP technology be very fast and operate on multiple channels simultaneously, rather than parallel processing in nature. In addition, there is a speed / power switch.

An alternative technique is the auxiliary threshold analog VLSI (C. Mead, Analog VLS
I and Nervous System, Addison-Wesley, 1989). This technology works at very low power levels and enables highly parallel processing circuits that operate at the low power utilized. In addition, one important property of the building block, communication capacity amplification, reflects the properties of biological systems that are better than digital on / off switches or linear analog instruments.

The sound is detected through a microphone. Alternatively, a direct silicon transducer for the pressure wave is used. In a preferred embodiment, there are two microphones, which are worn in the user's ear (either behind the ear or in the inner ear). The microphone is
It is multidirectional and needs to receive signals from all directions, so that the direction of the sound source can be estimated.

We describe below one possible neural embodiment of the several processes described above, which is given by way of example only and limits the inventive concept. do not do. This processing is performed in a plurality of stages. For the input from each ear, the first stage (after transformation) is cochlear filtering, (on any bandpassed channel), intensity calculation (parallelized), pitch phase detection and envelope processing ( This is the amplitude modulation phase detector and the rising detector). The result of this process (all channels, both ears) is used to generate an ITD estimate for either channel, with any characteristic type. This information is used to determine what should be presented to the user.

A neural technique for real-time cochlear filtering was first proposed by Rion et al. (RF Rion and C. Mead, Analog Electrical Cochlea, IEEE Acoustic Transformation, Speech and Signal Processing, 36 ( 7): 1119-1134, published 1988), by Lazaro (J. Lazaro and C. Mead, Silicon Model of Pitch, Perceptual Processing of the American Academy of Sciences, 86 (23): 9597-).
9601, 1989), by Liu and Androw (W. Liu, A. et al.
G. FIG. Androw and J. M. H. Goldstein, Analog Cochlear Model for Complex Decomposition Analysis, Advances in Neural Information Processing 5, 666-67
3, published in 1993; Liu, A. G. FIG. Androw and J. M. H.
Goldstein, voice conversation expression by analog silicon model around hearing,
IEEE Transformed Neural Network, 3 (3): 477-487, 1993), by Watt (L. Watt, Cochlea structure: analysis and analog VLSI P
hD, California Institute of Technology, 1993), and more recently,
Extended by Frazier and Vanshaik (EA Fragnier, A. Vanshaik and EA Vitos, Design of an Analog VLSI Model of the Active Cochlea, Analog Integrated Circuits and Signal Processing, 12: 19-35, 1997). Was done. The advantages of neural analysis are, unlike the DSP approach, essentially real-time and low power. At present, it is not possible to achieve high characteristic variables (Q) or many stages as achieved by the human cochlea, but it is still impossible to achieve the most recent technology (A. Vanshaik, Electrical Analog VLSI construction for auditory pathways, PhD theory, Lausanne's Association of Ecological Multi-Technologies, 1997) yielded 104 stages using second-order low-pass filter serialization.

[0095] Animal pitch phase detection relies on population coding by stimulating nerves to spike at the characteristic phase of periosteal movement of the skull. This neural approach has been discussed by Liu et al. (W. Liu, AG Androw and Jr. MH Goldstein, Spoken Speech Representation with Analog Silicon Model of Auditory Perception, I.
EEE conversion neural network, 3 (3): 477-487, published in 1993)
), In technology, Vanshaik (A. Vanshaik, Analog VLSI construction for electrical auditory pathways, PhD theory, Lausanne's Ecological Multi-Technological Union, 199
7 years), versions of the cilia cell model of Medis (MJ Hewitt and R. Medis, evaluation of eight computational models of mammalian internal cilia cell function, Journal of the American Acoustical Society, 90 (2): 904-917. Page, 1991). In both of these cases, the tendency to synchronize to an input signal below about 4 kHz and a quick, short-term adaptation are modeled.

However, if the goal is simply to encode the phase of the signal leaving each bandpass filter, a further application technique would be towards rectification and subsequent peak detection, or alternatively a simple positive. It was the detection of zero crossing. All of these are readily achievable with the use of neural morphology technology. (J. Lazaro and C. Mead, Silicon Model of Auditory Position Detection, Neural Computation Method, 1 (1): pp. 47-57, 1989). The neural morphological model used for zero crossings derived from the wavelength matched bandpass filter output was supplemented.

[0097] Even though the neural morphology system can maintain waveform-synchronous operation at high frequencies, sound source detection is difficult due to the short time of these signals. The peak match is
Ambiguous in the direction of the sound source. However, if the result of bandpassing the signal at the high frequency is amplitude modulation at the low frequency, then the difference in modulation phase between the two detectors can be used. Amplitude-modulated neural morphology detection (modeling cochlear nerve radial cells) was discussed by Van Shaik in the context of a periodical excerpt (A. Van Shaik, Analog VLSI Building Structure for Electrical Auditory Pathways, PhD theory, Lausanne's Union of Ecological Multi-Technologies, 1997).

The same technique can be used in ITD estimation, but applying a low-pass filter to the waveform-locked phase detector to produce a pulse at each peak (or at each positive zero crossing): Probably easy.

[0099] The neural morphology approach to rising detection can be achieved by using neuronal spiking neurons.

For each channel, there are three independent techniques for the ITD calculation method (amplitude modulation is not used below about 1 kHz, and waveform synchronization is about 4 kHz.
(Not used above), having many estimates in different parts of the spectrum, and even each part of the spectrum is inevitable to have many estimates.
If there are many sources at once, all of those estimates are correct.

A mix of sub-threshold analog, upper threshold analog and digital techniques is applied to control signal generation and generation of the blender's neural morphology approach.

What is to be represented is generated from the output data signal (or, in the case of binaural representation, the output data L signal and the output data R signal). Hearing technology, for example, remote generation of signals and transmission of signals to an in-the-ear transducer, wireless technology is available. In addition, there is a need to adjust the spectral energy distribution and make better use of the residual hearing.

Noting the bandpass characteristics of current neural morphological filters is that (i) selectively amplifying these channels so that the chosen ITD represents the strongest; and (ii) the chosen ITD By removing the contents of these channels, which are poorly represented, they are not as sharp as they can be balanced.

The embodiments of the present invention described above are given only by way of example and do not limit the concept in any way.

[Procedural Amendment] Submission of translation of Article 34 Amendment of the Patent Cooperation Treaty

[Submission date] July 14, 2000 (2000.7.14)

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Claims

[Correction method] Change

[Correction contents]

[Claims]

──────────────────────────────────────────────────続 き Continuation of front page (81) Designated country EP (AT, BE, CH, CY, DE, DK, ES, FI, FR, GB, GR, IE, IT, LU, MC, NL, PT, SE ), OA (BF, BJ, CF, CG, CI, CM, GA, GN, GW, ML, MR, NE, SN, TD, TG), AP (GH, GM, KE, LS, MW, SD, SL, SZ, UG, ZW), EA (AM, AZ, BY, KG, KZ, MD, RU, TJ, TM), AE, AL, AM, AT, AU, AZ, BA, BB, BG, BR , BY, CA, CH, CN, CU, CZ, DE, DK, EE, ES, FI, GB, GD, GE, GH, GM, HR, HU, ID, IL, IN, IS , JP, KE, KG, KP, KR, KZ, LC, LK, LR, LS, LT, LU, LV, MD, MG, MK, MN, MW, MX, NO, NZ, PL, PT, RO, RU, SD, SE, SG, SI, SK, SL, TJ, TM, TR, TT, UA, UG, US, UZ, VN, YU, ZA, ZW

Claims (26)

[Claims] 1. A sound processing method comprising the following steps: detecting a sound at at least two detection positions separated from each other; and converting the detected sound to identify an angular relationship between each sound source and the detection position. Analyzing, selecting an angular relationship associated with a particular sound source, and processing the detected speech in response to the selection to enhance a sound flow associated with the particular sound source. 2. The method of claim 1, wherein the angular relationship between the sound sources is determined at least in part with respect to a time difference between sounds from each of the sound sources detected at the separation detection location. 3. An angular relationship between the sound sources is determined with respect to a time difference determined with respect to at least one feature of the detected speech, wherein the feature is selected from the group consisting of a waveform phase, an amplitude modulation phase, and a rise. Item 3. The method according to Item 2. 4. The method of claim 3, wherein the time difference is determined for a plurality of features of the detected speech. 5. The angular relationship between the sound sources is determined at least in part with respect to intensity differences between sounds from each sound source detected at the separation detection position.
The method described in. 6. The method according to claim 1, wherein the sound is detected at a position corresponding to the user's ear. 7. The method of claim 6, wherein the angular relationship between the sound sources is determined, at least in part, with respect to interaural time difference (ITD). 8. The method according to claim 6, wherein the angular relationship between the sound sources is determined at least in part with respect to the interaural intensity difference (IID). 9. The method according to claim 1, further comprising the following steps:
Selectively filtering the detected sound from the separated position into a plurality of channels,
Then, the feature of the sound of each channel from one position is compared with the feature of the sound from the corresponding channel at another position. 10. The method of claim 9, wherein an angular relationship between the sound sources is determined with respect to a time difference determined with respect to a waveform phase of the detected voice, and the time differences are aggregated by summing up their values. 11. The method of claim 9, wherein an angular relationship between the sound sources is determined with respect to a time difference determined with respect to a rise of the detected voice, and the rises are aggregated in one ear prior to the determination of the time difference. 12. The method of claim 1, wherein an angular relationship between the sound sources is determined with respect to a time difference determined with respect to an amplitude modulation phase of the detected sound, and the amplitude modulation channels are aggregated at an amplitude modulation frequency before the time difference is determined. 9. The method according to 9. 13. A method for processing sounds from a plurality of sound sources, comprising:
Detecting sounds at at least two separated detection positions; a difference between an intensity difference between sounds from the sound sources detected at the separation detection position and a sound from each sound source detected at the separation detection position; Analyzing the detected sound to identify an angular relationship between each sound source and the detection location, and based on the determined angular relationship, a sound associated with at least one of the sound sources. Get the flow. 14. A sound processing apparatus comprising: a means for detecting sound at at least two detection positions separated from each other; and detecting the sound in order to identify an angular relationship between each sound source and the detection position. Means for analyzing, means for selecting an identified angular relationship associated with a particular sound source, and means for processing the detected speech in response to the selection to enhance a sound flow associated with the particular sound source. 15. The method according to claim 14, wherein the analyzing means includes means for determining a time difference between sounds from the sound sources detected at the separation detection position to determine an angular relationship between the sound sources. apparatus. 16. The method according to claim 16, wherein the analyzing unit determines a time difference between sounds from each of the sound sources detected at the separation detection position with respect to at least one of a waveform phase, an amplitude modulation phase, and a rise. And means for determining the angular relationship of
An apparatus according to claim 5. 17. An angle relationship between sound sources is determined by determining a time difference between sounds from the sound sources detected at the separation detection position with respect to a plurality of features of the detected sound. 17. The device of claim 16, including means. 18. The apparatus according to claim 13, wherein the analyzing means includes means for determining an angular relationship between the sound sources by determining an intensity difference between sounds from the sound sources detected at the separation detection position. The device according to any one of claims 17 to 17. 19. The apparatus according to claim 13, which is a hearing aid, and wherein the voice detecting means can be arranged at a position corresponding to the user's ear. 20. An analysis means, comprising: an interaural time difference (ITD) between sounds from the respective sound sources.
2.) means for determining the angular relationship between each sound source by determining
Device according to claim 9. 21. An analysis means, comprising: an interaural intensity difference (IID) between sounds from the respective sound sources.
2.) means for determining the angular relationship between each sound source by determining
21. The device according to 9 or 20. 22. The apparatus according to claim 14, further comprising: means for selectively filtering a detected sound from the separated position into a plurality of channels,
The analyzing means has means for comparing the characteristics of the sound of each channel from one position with the characteristics of the sound from the corresponding channel at another position. 23. The apparatus of claim 22, wherein an angular relationship between the sound sources is determined with respect to a time difference determined with respect to a waveform phase of the detected sound, and the time differences are aggregated by summing up their values. 24. The apparatus of claim 22, wherein an angular relationship between the sound sources is determined with respect to a time difference determined with respect to a rise of the detected sound, and the rises are aggregated with one ear prior to the determination of the time difference. 25. The angular relationship between the sound sources is determined with respect to a time difference determined with respect to an amplitude modulation phase of the detected voice, and the amplitude modulation channels are aggregated at an amplitude modulation frequency before the time difference is determined. 23. The method of claim 22. 26. A sound processing apparatus comprising: a means for detecting sound at at least two separated detection positions; an intensity difference between sounds from respective sound sources detected at the separated detection positions; Means for analyzing the detected sound to identify an angular relationship between each sound source and the detected position, with respect to at least one of sounds from each sound source detected at the separation detection position; and Means for obtaining a sound flow associated with at least one of the sound sources based on the determined angular relationship.