вход по аккаунту


код для вставкиСкачать
Patent Translate
Powered by EPO and Google
This translation is machine-generated. It cannot be guaranteed that it is intelligible, accurate,
complete, reliable or fit for specific purposes. Critical decisions, such as commercially relevant or
financial decisions, should not be based on machine-translation output.
An ultrasound Doppler technique is used to improve the quality of acoustic signals. The method
and system improve the quality of the acoustic signal acquired by the microphone from the
acoustic source while simultaneously acquiring the ultrasound Doppler signal from the moving
parts of the acoustic source. Then, by analyzing the acoustic signal and the Doppler signal
according to the model, an acoustic signal with improved quality is generated. [Selected figure]
Figure 1
Method of improving the quality of a noisy acoustic signal and system for acquiring the acoustic
signal to improve the quality of the acoustic signal
The present invention relates to signal processing, and in particular to acquiring and improving
acoustic signals.
There are many applications where acoustic signals are acquired by far-field microphones, such
as hands-free mobile communications, telephone and hands-free speech recognition.
In such applications, the acquired acoustic signal often contains a large amount of noise, such as
traffic, crowds, radio, TV, wind, or other ambient noise in the environment. Noise can be an
obstacle, especially in speech recognition, when interpreting acquired acoustic signals or when
performing other processing.
Doppler technology has been used for many applications. U.S. Patent No. 6,251,077 entitled
"Method and Apparatus for Dynamic Noise Reduction for Doppler Audio Output" issued to Mo et
al. On June 26, 2001 is for adaptive noise reduction. A method of suppressing background noise
of a spectral Doppler image using a low pass filter is described. US Patent No. 6,773,400 entitled
"Noninvasive transcranial Doppler ultrasound face and object recognition testing system", issued
to Njemanze on August 10, 2004, provides baseline blood flow in the cerebral arteries. A method
is described for testing a subject using face and object recognition tasks while measuring
velocity. U.S. Pat. No. 6,773,403, issued Aug. 10, 2004 to "Kim et al." Entitled "Ultra-sonic
apparatus and method for measuring the degrees of human tissue using the Doppler effects" We
describe a method of measuring the velocity of tissue components by sampling echo signals
reflected from the human body by generating a frequency distribution of data including velocity.
It is desirable to use ultrasound Doppler techniques to improve the quality of the acoustic signal.
SUMMARY OF THE INVENTION The present invention includes a Doppler ultrasound signal
generator, an acoustic signal detector, eg, a microphone, and an ultrasound signal sensor.
It is also possible to obtain both acoustic and ultrasound signals using a broadband microphone.
The ultrasound signal is higher in frequency than the nominal audio frequency range, for
example higher than 20 KHz.
In operation, a human speaker speaks while facing an acoustic detector and an ultrasonic sensor.
The detectors and sensors may be located on a desk, on a platform, or otherwise attached to, for
example, a dashboard or rearview mirror of a car. The acoustic microphone acquires an acoustic
signal generated by the speaker.
At the same time, the ultrasound generator transmits high frequency acoustic signals towards the
face of the speaker. The ultrasound signal is reflected by the area around the speaker's face, in
particular the mouth, ie the lips, the tongue and the jaws. The reflected Doppler signal is acquired
by the ultrasonic sensor. The frequency of the reflected Doppler signal is modulated by the
movement of the speaker's face, tongue, lips and mouth as the acoustic signal is generated.
Thus, the acoustic and Doppler signals are very correlated. The acquired ultrasound signal is
analyzed in conjunction with the acoustic signal to improve the quality of the acoustic signal. The
enhanced quality acoustic signal can then be further processed in a number of applications, such
as in an improved speech recognition system.
System Configuration FIG. 1 shows a system 100 that uses ultrasound signals to improve the
quality of noisy acoustic signals. An acoustic signal is defined herein as being nominally less than
20 KHz, ie the acoustic signal is a signal that can be heard by the human ear, and the ultrasound
signal is a signal that is higher in frequency than the acoustic signal. That is, the frequencies of
the acoustic signal and the ultrasonic signal do not have the same element.
System 100 includes a housing 110 incorporating an acoustic microphone 102, an ultrasound
transducer 103 and an ultrasound sensor 104. Instead of acoustic microphones and ultrasound
sensors, broadband microphones that can be sensed over a wide range of acoustic frequencies,
including ultrasound Doppler frequencies, may be used.
The acoustic microphone 102 obtains a noisy acoustic signal 105 from an acoustic source 101,
for example a human speaker. The noisy acoustic signal is converted into an electrical signal 106
representing the acoustic signal, which is detected 120. It should be noted that the sound source
may be any other sound source, such as a speaker cone or a diaphragm or a machine with
moving parts. In this case, the invention can determine when the machine is operating properly
by analyzing together the acoustic signal and the reflected Doppler signal modulated by the
movable part.
The ultrasonic signal generator 130 generates an ultrasonic signal 108 to the transducer 103.
The ultrasound signal is generally directed to the acoustic source 101 and the reflected Doppler
signal 109 is separately detected 140.
When an acoustic signal is generated by a moving part of the acoustic source, for example the
mouth, the lip and the tongue, the reflected Doppler signal is highly correlated to the
corresponding acoustic signal.
Thus, the invention uses a model that represents the state of the "clean" acoustic spectrum and
the corresponding Doppler spectrum.
This model can then be used to improve the quality of the inherently noisy acoustic signal by
correlating the acquired Doppler signal with the corresponding clean acoustic signal.
By combining the detected noisy acoustic signal and the Doppler signal according to the model
200 and analyzing it, a quality enhanced acoustic signal 151 with a reduced amount of noise is
generated 150. The enhanced quality acoustic signal 151 can be further processed, for example,
to perform speech recognition 160.
Mixed Model Training FIG. 2 shows a model 200 that uses ultrasound signals to improve the
quality of acoustic signals such as speech. This model is trained using an acoustic spectrum 201,
a Doppler or ultrasound spectrum 202 and a noise spectrum 203. The noise spectrum 203 is
added 210 to the acoustic spectrum 201 to provide a noisy acoustic spectrum 204.
The model has a plurality of states 220. In each state, there is one spectrum 201 for the "clean"
acoustic signal and one spectrum 202 for the corresponding Doppler signal. Such a distribution,
It may be expressed mathematically as
Here, z represents the state, D represents the Doppler spectrum, and S represents, for example,
the acoustic spectrum for a 30 millisecond segment of the acquired signal.
The parameters of the model include the prior probability of the state P (z) of the model 200, the
state dependent distribution P (D ¦ z) of the Doppler spectrum, and the state dependent
distribution P (S ¦ z) of the acoustic spectrum. The model is "trained" from a corpus of
simultaneous "clean" acoustic signals and corresponding Doppler signals. The model may be in
the form of a mixture of Gaussian distributions, each distribution having a mean and a variance.
It is also possible to use other models, such as Hidden Markov Models (HMMs) or Bayesian
According to the model, the acoustic signal generation process or the acoustic source, eg face or
machine, is in a different state at every moment.
In that state, the acoustic source produces a single spectrum for the acoustic signal and a
corresponding single spectrum for the Doppler signal that can be correlated with the acoustic
signal simultaneously. The acoustic spectrum 201 is contaminated by the additive noise 203 to
generate a noisy acoustic spectrum 204.
Estimating a Quality-Enhanced Acoustic Signal from a Noise-Containing Acoustic Signal FIG. 3
illustrates a method 300 for obtaining an enhanced quality acoustic signal 151 estimated from a
noisy acoustic signal 105 using a Doppler signal. . The high frequency Doppler spectrum 302 is
not contaminated by lower frequency acoustic signals. The Doppler spectrum provides the
"evidence" used to estimate the posterior probabilities of the various states 220 of the model
200. These estimates are considered to be reliable as the Doppler signal is not contaminated by
acoustic noise.
A model for the noise spectrum is used to obtain a state-dependent estimate of the enhanced
acoustic spectrum 301 from the noisy acoustic signal 105 in an analysis and generation step
310. The noise spectrum and the Doppler spectrum are combined using the posterior probability
of the state obtained from the Doppler signal as a weight to produce an estimate for the quality
improved acoustic signal 151.
An example procedure can be described as follows. Y represents the acoustic spectrum of speech
containing noise. Let f (S, N) denote a function representing the effect of noise N on the acoustic
spectrum S of the clean acoustic signal. すなわち、Y=f(S,N)とする。 The noise is
unknown. N is not known.
Ideally, the noise can be estimated through the inverse function N = f <1> (Y, S) if the clean
acoustic spectrum S is also known. Unfortunately, the clean speech spectrum S is also unknown.
However, if the state z that generated the acoustic spectrum S is known, then the noise N
It can be estimated as
Here, μz is an average value of the state dependent distribution P (S ¦ z) of the acoustic
spectrum. However, the state z is also unknown and can not be accurately estimated from the
acquired acoustic signal because the acoustic signal contains noise. However, the Doppler
spectrum D is not contaminated by noise and by deriving evidence from the Doppler spectrum
The posterior probability of state z can be estimated as
The estimated spectrum of noise can then be obtained as a weighted combination of estimates
derived from each of the states.
Where the weight of the state is
Is obtained from the Doppler spectrum according to.
Inverse transform using estimated noise
Through the
noise removal
of the spectrum of the acquired acoustic signal.
This transformation represents how noise removal can be performed on the noisy acoustic signal
according to the invention.
It should be understood that the transformations described above are only an example.
Depending on the procedure, various transformation functions f (.
And g (.
) Can be used.
Similarly, other statistical models can be used to represent different spectra.
The present invention allows many applications, such as speech recognition, to operate with
improved quality acoustic signals. For example, using the present invention to remove signal
noise from hands free phones in a car and to improve the quality of the transmitted signal in a
mobile phone when it is used in a noisy environment It can be used to remove noise or diagnose
the operation of machinery and detect potential failures.
Although the invention has been described by way of examples of preferred embodiments, it is to
be understood that various other adaptations and modifications may be made within the spirit
and scope of the invention. Therefore, it is the object of the appended claims to cover all such
variations and modifications as come within the true spirit and scope of the present invention.
FIG. 1 is a block diagram of a system and method for acquiring an acoustic signal according to
the present invention. Fig. 2 is a block diagram of a model according to the invention. FIG. 5 is a
block diagram of a method of improving the quality of an acoustic signal according to the
present invention.
Пожаловаться на содержимое документа