EP1010169B1 - Channel gain modification system and method for noise reduction in voice communication - Google Patents
Channel gain modification system and method for noise reduction in voice communication Download PDFInfo
- Publication number
- EP1010169B1 EP1010169B1 EP97945400A EP97945400A EP1010169B1 EP 1010169 B1 EP1010169 B1 EP 1010169B1 EP 97945400 A EP97945400 A EP 97945400A EP 97945400 A EP97945400 A EP 97945400A EP 1010169 B1 EP1010169 B1 EP 1010169B1
- Authority
- EP
- European Patent Office
- Prior art keywords
- noise
- speech
- gain
- audio signal
- channel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Lifetime
Links
- 238000000034 method Methods 0.000 title claims description 44
- 238000004891 communication Methods 0.000 title description 11
- 238000012986 modification Methods 0.000 title description 2
- 230000004048 modification Effects 0.000 title description 2
- 230000005236 sound signal Effects 0.000 claims description 38
- 230000001419 dependent effect Effects 0.000 claims description 4
- 238000005311 autocorrelation function Methods 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims 4
- 238000012886 linear function Methods 0.000 claims 2
- 230000001629 suppression Effects 0.000 description 29
- 238000012545 processing Methods 0.000 description 19
- 230000003595 spectral effect Effects 0.000 description 8
- 230000007613 environmental effect Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 230000005540 biological transmission Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000001413 cellular effect Effects 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 238000012805 post-processing Methods 0.000 description 3
- 230000002238 attenuated effect Effects 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000737 periodic effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000010267 cellular communication Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
- G10L21/0232—Processing in the frequency domain
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/04—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using predictive techniques
- G10L19/16—Vocoder architecture
- G10L19/18—Vocoders using multiple modes
- G10L19/24—Variable rate codecs, e.g. for generating different qualities using a scalable representation such as hierarchical encoding or layered encoding
Definitions
- the present invention relates to speech processing. More particularly, the present invention relates to a noise suppression system and method for use in speech processing.
- Noise suppression in a speech communication system generally serves the purpose of improving the overall quality of the desired audio signal by filtering environmental background noise from the desired speech signal.
- This speech enhancement process is particularly necessary in environments having abnormally high levels of ambient background noise, such as an aircraft, a moving vehicle, or a noisy factory.
- One noise suppression technique is the spectral subtraction, or spectral gain modification, technique.
- the input audio signal is divided into frequency channels, and particular frequency channels are attenuated according to their noise energy content.
- a background noise estimate for each frequency channel is utilized to generate a signal-to-noise ratio (SNR) of the speech in the channel, and the SNR is used to compute a gain factor for each channel.
- the gain factor determines the attenuation for the particular channel.
- the attenuated channels are recombined to produce the noise-suppressed output signal.
- the speakerphone option provides hands-free operation for the automobile driver.
- the hands-free microphone is typically located at a greater distance from the user, such as being mounted overhead on the visor.
- the distant microphone delivers a poor SNR to the land-end party due to road and wind noise conditions.
- the received speech at the land-end is usually intelligible, continuous exposure to such background noise levels often increases listener fatigue.
- Spectral subtraction techniques update the background noise estimate during periods when speech is absent. When speech is absent, the measured spectral energy is attributed to noise, and the noise estimate is updated based on the measured spectral energy. Therefore, it is important to distinguish between periods of speech and absence of speech in order to obtain an accurate noise energy estimate for computation of the SNR.
- An exemplary technique for speech detection uses a voice metric calculator to perform the noise update decision.
- a voice metric is a measurement of the overall voice-like characteristics of the channel energy.
- raw SNR estimates are used to index a voice metric table to obtain voice metric values for each channel.
- the individual channel voice metric values are summed to create an energy parameter, which is compared with a background noise update threshold. If the voice metric sum meets or exceeds the threshold, then the signal is said to contain speech. If the voice metric sum does not meet the threshold, the input frame is deemed to be noise, and a background noise update is performed.
- SNR measurements will be large, resulting in a high voice metric, which negates a noise estimate update.
- a refinement to the voice metric calculator technique measures the channel energy deviation. This method assumes that noise exhibits constant spectral energy over time, while speech exhibits variable spectral energy over time. Thus, the channel energy is integrated over time, and speech is detected if there is substantial channel energy deviation, while noise is detected if there is little channel energy deviation.
- a speech detector which measures channel energy deviation will detect a sudden increase in the level of noise.
- the channel energy deviation method provides an inaccurate result when the input speech signal is of constant energy.
- changes in the input energy will cause the energy deviation to be large, negating a noise estimate update even though an update is necessary.
- the noise suppression system In addition to an accurate speech detector, the noise suppression system must appropriately adjust channel gains. Channel gains should be adjusted so that noise suppression is achieved without sacrificing the voice quality.
- One method of channel gain adjustment computes the gain as a function of the total noise estimate and the SNR of the speech signal. In general, an increase in the total noise estimate results in a lower gain factor for a given SNR. A lower gain factor is indicative of a greater attenuation factor. This technique imposes a minimum gain value to prevent excess attenuation of the channel gain when the total noise estimate is very high.
- a hard clamped minimum gain value a tradeoff between noise suppression and voice quality is introduced. When the clamp is relatively low, noise suppression is improved but voice quality is degraded. When the clamp is relatively high, noise suppression is degraded but the voice quality is improved.
- US Patent No. 4, 811, 404 "Motorola Inc.” discloses a method and apparatus for suppressing background noise in high background noise environments.
- This US Patent includes the addition of a signal to noise ratio (SNR) threshold mechanism to reduce background noise flutter by off-setting the gain rise of a gain table until a certain SNR threshold is reached.
- SNR signal to noise ratio
- the present invention as set out in the appended claims is a noise suppression system and method for use in speech processing systems.
- An objective of the present invention is to provide a speech detector which determines the presence of speech in an input signal.
- a reliable speech detector is needed for an accurate determination of the signal-to-noise ratio (SNR) of speech.
- SNR signal-to-noise ratio
- the input signal is assumed to be entirely a noise signal, and the noise energy may be measured.
- the noise energy is then used for determination of the SNR.
- Another objective of the present invention is to provide an improved gain determination element for realization of noise suppression.
- the noise suppression system comprises a speech detector which determines if speech is present in a frame of the input signal.
- the speech decision may be based on the SNR measure of speech in an input signal.
- a SNR estimator estimates the SNR based on the signal energy estimate generated by an energy estimator and the noise energy estimate generated by a noise energy estimator.
- the speech decision may also be based on the encoding rate of the input signal.
- each input frame is assigned an encoding rate selected from a predetermined set of rates based on the content of the input frame.
- the rate is dependent on the level of speech activity, so that a frame containing speech would be assigned -a high rate, whereas a frame not containing speech would be assigned a low rate.
- the speech decision may be based on one or more mode measures which are descriptive of the characteristics of the input signal. If it is determined that speech is not present in the input frame, then the noise energy estimator updates the noise energy estimate.
- a channel gain estimator determines the gain for the frame of input signal. If speech is not present in the frame, then the gain is set to be a predetermined minimum. Otherwise, the gain is determined based on the frequency content of the frame.
- a gain factor is determined for each of a set of predefined frequency channels. For each channel, the gain is determined in accordance with the SNR of the speech in the channel. For each channel, the gain is defined using a function that is suitable for the characteristics of the frequency band within which the channel is located. Typically, for a predefined frequency band, the gain is set to increase linearly with increasing SNR. Additionally, the minimum gain for each frequency band may be adjustable based on the environmental characteristics. For example, a user-selectable minimum gain may be implemented.
- the channel SNRs are based on channel energy estimates generated by an energy estimator and channel noise energy estimates generated by a noise energy estimator.
- the gain factors are used to adjust the gain of the signal in the different channels, and the gain adjusted channels are combined to produce the noise suppressed output signal.
- noise suppressors are commonly used to suppress undesirable environmental background noise.
- Most noise suppressors operate by estimating the background noise characteristics of the input data signal in one or more frequency bands and subtracting an average of the estimate(s) from the input signal. The estimate of the average background noise is updated during periods of the absence of speech.
- Noise suppressors require an accurate determination of the background noise level for proper operation.
- the level of noise suppression must be properly adjusted based on the speech and noise characteristics of the input signal.
- FIG. 1 An exemplary speech processing system 100 in which the present invention may be embodied is illustrated in FIG. 1 .
- System 100 comprises microphone 102 , A/D converter 104 , speech processor 106 , transmitter 110 , and antenna 112 .
- Microphone 102 may be located in a cellular telephone together with the other elements illustrated in FIG. 1 .
- microphone 102 may be the hands-free microphone of the vehicle speakerphone option to a cellular communication system.
- the vehicle speakerphone assembly is sometimes referred to as a carkit. Where microphone 102 is part of a carkit, the noise suppression function is particularly important. Because the hands-free microphone is generally positioned at some distance from the user, the received acoustic signal tends to have a poor speech SNR due to road and wind noise conditions.
- the input audio signal comprising speech and/or background noise
- the input audio signal is transformed by microphone 102 into an electro-acoustic signal represented by the term s(t).
- the electro-acoustic signal may be converted from an analog signal to pulse code modulated (PCM) samples by Analog-to-Digital converter 104 .
- PCM samples are output by A/D converter 104 at 64 kbps and are represented by signal s(n) as shown in FIG. 1 .
- Digital signal s(n) is received by speech processor 106 , which comprises, among other elements, noise suppressor 108 .
- Noise suppressor 108 suppresses noise in signal s(n) in accordance with the present invention.
- noise suppressor 108 determines the level of background environmental noise and adjusts the gain of the signal to mitigate the effects of such environmental noise.
- speech processor 106 generally comprises a voice coder, or a vocoder (not shown), which compresses speech by extracting parameters that relate to a model of human speech generation. Speech processor 106 may also comprise an echo canceller (not shown), which eliminates acoustic echo resulting from the feedback between a speaker (not shown) and microphone 102 .
- transmitter 110 Following processing by speech processor 106 , the signal is provided to transmitter 110 , which performs modulation in accordance with a predetermined format such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), or Frequency Division Multiple Access (FDMA).
- CDMA Code Division Multiple Access
- TDMA Time Division Multiple Access
- FDMA Frequency Division Multiple Access
- transmitter 110 modulates the signal in accordance with a CDMA modulation format as described in U.S. Pat. No. 4,901,307, entitled "SPREAD SPECTRUM MULTIPLE ACCESS COMMUNICATION SYSTEM USING SATELLITE OR TERRESTRIAL REPEATERS," which is assigned to the assignee of the present invention.
- Transmitter 110 then upconverts and amplifies the modulated signal, and the modulated signal is transmitted through antenna 112 .
- noise suppressor 108 may be embodied in speech processing systems that are not identical to system 100 of FIG. 1 .
- noise suppressor 108 may be utilized within an electronic mail application having a voice mail option.
- transmitter 110 and antenna 112 of FIG. 1 will not be necessary.
- the noise suppressed signal will be formatted by speech processor 106 for transmission through the electronic mail network.
- FIG. 2 An exemplary embodiment of noise suppressor 108 is illustrated in FIG. 2 .
- the input audio signal is received by preprocessor 202 , as shown in FIG. 2 .
- Preprocessor 202 prepares the input signal for noise suppression by performing preemphasis and frame generation. Preemphasis redistributes the power spectral density of the speech signal by emphasizing the high frequency speech components of the signal. Essentially performing a high pass filtering function, preemphasis emphasizes the important speech components to enhance the SNR of these components in the frequency domain.
- Preprocessor 202 may also generate frames from the samples of the input signal. In a preferred embodiment, 10 ms frames of 80 samples/frame are generated. The frames may have overlapped samples for- better processing accuracy.
- the frames may be generated by windowing and zero padding of the samples of the input signal.
- the preprocessed signal is presented to transform element 204 .
- transform element 204 generates a 128 point Fast Fourier Transform (FFT) for each frame of input signal. It should be understood, however, that alternative schemes may be used to analyze the frequency components of the input signal.
- FFT Fast Fourier Transform
- the transformed components are provided to channel energy estimator 206a , which generates an energy estimate for each of N channels of the transformed signal.
- the low frequency channel corresponds to frequency range from 250 to 2250 Hz
- the high frequency channel corresponds to frequency range from 2250 to 3500 Hz.
- the current channel energy of the low frequency channel may be determined by summing the energy of the FFT points corresponding to 250-2250 Hz
- the current channel energy of the high frequency channel may be determined by summing the energy of the FFT points corresponding to 2250-3500 Hz.
- the energy estimates are provided to speech detector 208 , which determines whether or not speech is present in the received audio signal.
- SNR estimator 210a of speech detector 208 receives the energy estimates.
- SNR estimator 210a determines the signal-to-noise ratio (SNR) of the speech in each of the N channels based on the channel energy estimates and the channel noise energy estimates.
- the channel noise energy estimates are provided by noise energy estimator 214a , and generally correspond to the estimated noise energy smoothed over the previous frames which do not contain speech.
- Speech detector 208 also comprises rate decision element 212 , which selects the data rate of the input signal from a predetermined set of data rates.
- data is encoded so that the data rate may be varied from one frame to another. This is known as a variable rate communication system.
- the voice coder which encodes data based on a variable rate scheme is typically called a variable rate vocoder.
- An exemplary embodiment of a variable rate vocoder is described in U.S. Pat. No. 5,414,796, entitled "VARIABLE RATE VOCODER,” assigned to the assignee of the present invention.
- the use of a variable rate communications channel eliminates unnecessary transmissions when there is no useful speech to be transmitted.
- Algorithms are utilized within the vocoder for generating a varying number of information bits in each frame in accordance with variations in speech activity. For example, a vocoder with a set of four rates may produce 20 millisecond data frames containing 16, 40, 80, or 171 information bits, depending on the activity of the speaker. It is desired to transmit each data frame in a fixed amount of time by varying the transmission rate of communications.
- determining the rate will provide information on whether speech is present or not.
- a determination that a frame should be encoded at the highest rate generally indicates the presence of speech, while a determination that a frame should be encoded at the lowest rate generally indicates the absence of speech.
- Intermediate rates typically indicate transitions between the presence and the absence of speech.
- Rate decision element 212 may implement any of a number of rate decision algorithms.
- One such rate decision algorithm is disclosed in copending U.S. patent application Serial No. 08/286,842, entitled “METHOD AND APPARATUS FOR PERFORMING REDUCED RATE VARIABLE RATE VOCODING,” assigned to the assignee of the present invention.
- This technique provides a set of rate decision criteria referred to as mode measures.
- a first mode measure is the target matching signal to noise ratio (TMSNR) from the previous encoding frame, which provides information on how well the encoding model is performing by comparing a synthesized speech signal with the input speech signal.
- a second mode measure is the normalized autocorrelation function (NACF), which measures periodicity in the speech frame.
- TMSNR target matching signal to noise ratio
- NACF normalized autocorrelation function
- a third mode measure is the zero crossings (ZC) parameter, which measures high frequency content in an input speech frame.
- ZC zero crossings
- PPD prediction gain differential
- ED energy differential
- rate decision element 212 is shown in FIG. 2 as an included element of noise suppressor 108
- the rate information may instead be provided to noise suppressor 108 by another component of speech processor 106 (FIG. 1 ).
- speech processor 106 may comprise a variable rate vocoder (not shown) which determines the encoding rate for each frame of input signal.
- the rate information may be provided to noise suppressor 108 by the variable rate vocoder.
- speech detector 208 may use a subset of the mode measures that contribute to the rate decision.
- rate decision element 212 may be substituted by a NACF element (not shown), which, as explained earlier, measures periodicity in the speech frame.
- N refers to the numbers of samples of the speech frame
- t1 and t2 refer to the boundaries within the T samples for which the NACF is evaluated.
- the NACF is evaluated based on the formant residual signal, e(n).
- Formant frequencies are the resonance frequencies of speech.
- a short term filter is used to filter the speech signal to obtain the formant frequencies.
- the residual signal obtained after filtering by the short term filter is the formant residual signal, and contains the long term speech information, such as the pitch, of the signal.
- the NACF mode measure is suitable for determining the presence of speech because the periodicity of a signal containing voiced speech is different from a signal which does not contain voiced speech.
- a voiced speech signal tends to be characterized by periodic components.When voiced speech is not present, the signal generally will not have periodic components.
- the NACF measure is a good indicator which may be used by speech detector 208 .
- Speech detector 208 may use measures such as the NACF instead of the rate decision in situations where it is not practicable to generate the rate decision. For example, if the rate decision is not available from the variable rate vocoder, and noise processor 108 does not have the processing power to generate its own rate decision, then mode measures like the NACF offer a desirable alternative. This may be the case in a carkit application where processing power is generally limited.
- speech detector 208 may make a determination regarding the presence of speech based on the rate decision, the mode measure(s), or the SNR estimate alone. Although additional measures should improve the accuracy of the determination, any one of the measures alone may provide an adequate result.
- the rate decision (or the mode measure(s)) and the SNR estimate generated by SNR estimator 210a are provided to speech decision element 216 .
- Speech decision element 216 generates a decision on whether or not speech is present in the input signal based on its inputs. The decision on the presence of speech will determine if a noise energy estimate update should be performed. The noise energy estimate is used by SNR estimator 210a to determine the SNR of the speech in the input signal. The SNR will in turn will be used to compute the level of attenuation of the input signal for noise suppression. If it is determined that speech is present, then speech decision element 216 opens switch 218a , preventing noise energy estimator 214a from updating the noise energy estimate.
- the input signal is assumed to be noise, and speech decision element 216 closes switch 218a , causing noise energy estimator 214a to update the noise estimate.
- switch 218a it should be understood that an enable signal provided by speech decision element 216 to noise energy estimator 214a may perform the same function.
- speech decision element 216 generates the noise update decision based on the procedure below:
- the channel SNR estimates provided by SNR estimator 210a are denoted by chsnr1 and chsnr2.
- the rate of the input signal, provided by rate decision element 212 is denoted by rate.
- a counter, ratecount keeps track of the number of frames based on certain conditions as described below.
- Speech decision element 216 determines that speech is not present, and that the noise estimate should be updated, if the rate is the minimum rate of the variable rates, either chsnr1 is greater than threshold T1 or chsnr2 is greater than threshold T2, and ratecount is greater than threshold T3. If the rate is minimum, and either chsnr1 is greater than T1 or chsnr2 is greater than T2, but ratecount is less than T3, then the ratecount is increased by one but no noise estimate update is performed.
- the counter, ratecount detects the case of a sudden increased level of noise or an increasing noise source by counting the number of frames having minimum rate but also having high energy in at least one of the channels.
- the counter which provides an indicator that the high SNR signal contains no speech, is set to count until speech is detected in the signal.
- ratecount is reset to zero.
- speech decision element 216 will determine that the frame contains speech, and no noise estimate update is performed, but ratecount is reset to zero.
- Speech decision element 216 may make use of the NACF measure to determine the presence of speech; and thus the noise update decision, in accordance with the procedure below: where pitchPresent is defined as follows: Again, channel SNR estimates provided by SNR estimator 210a are denoted by chsnr1 and chsnr2. A NACF element (not shown) generates a measure indicative of the presence of pitch, pitchPresent, as defined above. A counter, pitchCount, keeps track of the number of frames based on certain conditions as described below.
- the measure pitchPresent determines that pitch is present if NACF is above threshold TT1. If NACF falls within a mid range (TT2 ⁇ NACF ⁇ TT1) for a number of frames greater than threshold TT3, then pitch is also determined to be present.
- the counter, pitchCount is used to detect the case of a sudden increased level of noise or an increasing noise source.
- pitchPresent indicates that pitch is not present, and chsnr1 is less than TH1 and chsnr2 is less than TH2, then speech decision element 216 will determine that speech is not present and that a noise estimate update should be performed. In addition, pitchCount is reset to zero.
- speech decision element 216 will determine that the frame contains speech, and no noise estimate update is performed. However, pitchCount is reset to zero.
- Noise energy estimator 214a Upon determination that speech is not present, switch 218a is closed, causing noise energy estimator 214a to update the noise estimate.
- Noise energy estimator 214a generally generates a noise energy estimate for each of the N channels of the input signal. Since speech is not present, the energy is presumed to be wholly contributed by noise. For each channel, the noise energy update is estimated to be the current channel energy smoothed over channel energies of previous frames which do not contain speech.
- the updated channel noise energy estimates are presented to SNR estimator 210a . These channel noise energy estimates will be used to obtain channel SNR estimate updates for the next frame of input signal.
- Channel gain estimator 220 determines the gain, and thus the level of noise suppression, for the frame of input signal. If speech decision element 216 has determined that speech is not present, then the gain for the frame is set at a predetermined minimum gain level. Otherwise, the gain is determined as a function of frequency. In a preferred embodiment, the gain is computed based on the graph shown in FIG. 3. Although shown in graphical form in FIG. 3 , it should be understood that the function illustrated in FIG. 3 may be implemented as a look-up table in channel gain estimator 220 .
- a preferred embodiment of the present invention defines a separate gain curve for each of L frequency bands.
- the gain factor for a channel in the low band may be determined using the low band curve
- the gain factor for a channel in the mid band may be determined using the mid band curve
- the gain factor for a channel in the high band may be determined using the high band curve.
- environmental noise such as road and wind noise
- the energy of the noise signal is greater at the lower frequencies, and the energy generally decreases with increasing frequency.
- the preferred embodiment assigns the low band as 125 - 375 Hz, the mid band as 375 - 2625 Hz, and the high band as 2625 - 4000 Hz.
- the slopes and the y intercepts are experimentally determined.
- the preferred embodiment uses the same slope, 0.39, for each of the three bands, although a different slope may be used for each frequency band.
- lowBandYintercept is set at -17 dB
- midBandYintercept is set at -13 dB
- highBandYintercept is set at -13 dB.
- An optional feature would provide the user of the device comprising the noise suppressor to select the desired y-intercepts.
- more noise suppression (a lower y-intercept) may be chosen at the expense of some voice degradation.
- the y-intercepts may be variable as a function of some measure determined by noise suppressor 108 .
- more noise suppression (a lower y-intercept) may be desired when an excessive noise energy is detected for a predetermined period of time.
- less noise suppression (a high y-intercept) may be desired when a condition such as babble is detected. During a babble condition, background speakers are present, and less noise suppression may be warranted to prevent cut out of the main speaker.
- Another optional feature would provide for selectable slopes of the gain curves. Further, it should be understood that a curve other than the lines described by equations (4)-(6) may be found to be more suitable for determining the gain factor under certain circumstances.
- a gain factor is determined for each of M frequency channels of the input signal, where M is the predetermined number of channels to be evaluated.
- M is the predetermined number of channels to be evaluated.
- the channel SNR is used to derive the gain factor based on the appropriate curve.
- the channel SNRs are shown, in FIG. 2 , to be evaluated by channel energy estimator 206b , noise energy estimator 214b , and SNR estimator 210b .
- channel energy estimator 206b For each frame of input signal, channel energy estimator 206b generates energy estimates for each of M channels of the transformed input signal, and provides the energy estimates to SNR estimator 210b .
- the channel energy estimates may be updated using the relationship of Equation (1) above. If it is determined by speech decision element 216 that no speech is present in the input signal, then switch 218b is closed, and noise energy estimator 214b updates the estimates of the channel noise energy.
- the updated noise energy estimate is based on the channel energy estimate determined by channel energy estimator 206b .
- the updated estimate may be evaluated using the relationship of Equation (3) above.
- the channel noise estimates are provided to SNR estimator 210b .
- SNR estimator 210b determines channel SNR estimates for each frame of speech based on the channel energy estimates for the particular frame of speech and the channel noise energy estimates provided by noise energy estimator 214b .
- channel energy estimator 206a performs functions similar to channel energy estimator 206b , noise energy estimator 214b , switch 218b , and SNR estimator 210b , respectively.
- channel energy estimators 206a and 206b may be combined as one processing element
- noise energy estimators 214a and 214b may be combined as one processing element
- switches 218a and 218b may be combined as one processing element
- SNR estimators 210a and 210b may be combined as one processing element.
- the channel gain factors are provided by channel gain estimator 220 to gain adjuster 224 .
- Gain adjuster 224 also receives the FFT transformed input signal from transform element 204 .
- the gain adjusted signal generated by gain adjuster 224 is then provided to inverse transform element 226 , which in a preferred embodiment generates the Inverse Fast Fourier Transform (IFFT) of the signal.
- IFFT Inverse Fast Fourier Transform
- the inverse transformed signal is provided to post processing element 228 . If the frames of input had been formed with overlapped samples, then post processing element 228 adjusts the output signal for the overlap. Post processing element 228 also performs deemphasis if the signal had undergone preemphasis. Deemphasis attenuates the frequency components that were emphasized during preemphasis. The preemphasis/deemphasis process effectively contributes to noise suppression by reducing the noise components lying outside of the range of the processed frequency components.
- IFFT Inverse Fast Fourier Transform
- DSP digital signal processor
- ASIC application specific integrated circuit
- FIG. 4 a flow chart is shown illustrating some of the steps involved in the processing as discussed with reference to FIGS. 2 and 3 . Although shown as consecutive steps, one skilled in the art would recognize that ordering of some of the steps are interchangeable.
- transform element 204 transforms the input audio signal into a transformed signal, generally a FFT signal.
- SNR estimator 210b determines the speech SNR for M channels of the input signal based on the channel energy estimates provided by channel energy estimator 206b and the channel noise energy estimates provided by noise energy estimator 214b .
- channel gain estimator 220 determines gain factors for the M channels of the input signal based on the frequency of the channels. Channel gain estimator 220 sets the gain at a minimum level if speech has been found to be absent in the frame of input signal. Otherwise, a gain factor is determined, for each of the M channels, based on a predetermined function.
- a function defined by line equations having fixed slopes and y-intercepts, wherein each line equation defines the gain for a predetermined frequency band may be used.
- gain adjuster 224 adjusts the gain of the M channels of the transformed signal using the M gain factors.
- inverse transform element 226 inverse transforms the gain adjusted transformed signal, producing the noise suppressed audio signal.
- SNR estimator 210a determines the speech SNR for N channels of the input signal based on the channel energy estimates provided by channel energy estimator 206a and the channel noise energy estimates provided by noise energy estimator 214a .
- rate decision element 212 determines the encoding rate for the input signal through analysis of the input signal. Alternatively, one or more mode measures, such as the NACF, may be determined.
- speech decision element 216 determines if speech is present in the input signal based on the SNR provided by SNR estimator 210a, the rate provided by rate decision element 212 , and/or the mode measure(s).
- Noise energy estimator 214a updates the noise estimate based on the channel energy determined by channel energy estimator 206a. Whether or not speech is detected, the procedure continues to process the next frame of the input signal.
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Noise Elimination (AREA)
Description
- The present invention relates to speech processing. More particularly, the present invention relates to a noise suppression system and method for use in speech processing.
- Transmission of voice by digital techniques has become widespread, particularly in cellular telephone and personal communication system (PCS) applications. This, in turn, has created an interest in improving speech processing techniques. One area in which improvements are being developed is that of noise suppression techniques.
- Noise suppression in a speech communication system generally serves the purpose of improving the overall quality of the desired audio signal by filtering environmental background noise from the desired speech signal. This speech enhancement process is particularly necessary in environments having abnormally high levels of ambient background noise, such as an aircraft, a moving vehicle, or a noisy factory.
- One noise suppression technique is the spectral subtraction, or spectral gain modification, technique. Using this approach, the input audio signal is divided into frequency channels, and particular frequency channels are attenuated according to their noise energy content. A background noise estimate for each frequency channel is utilized to generate a signal-to-noise ratio (SNR) of the speech in the channel, and the SNR is used to compute a gain factor for each channel. The gain factor then determines the attenuation for the particular channel. The attenuated channels are recombined to produce the noise-suppressed output signal.
- In specialized applications involving relatively high background noise environments, most noise suppression techniques exhibit significant performance limitations. One example of such an application is the vehicle speakerphone option to a cellular mobile communication system. The speakerphone option provides hands-free operation for the automobile driver. The hands-free microphone is typically located at a greater distance from the user, such as being mounted overhead on the visor. The distant microphone delivers a poor SNR to the land-end party due to road and wind noise conditions. Although the received speech at the land-end is usually intelligible, continuous exposure to such background noise levels often increases listener fatigue.
- For a noise suppression system to function properly, it is important to accurately determine the SNR of speech. However, it is difficult to accurately determine the SNR for the speech signal because of the limitations of currently available noise detectors. Spectral subtraction techniques update the background noise estimate during periods when speech is absent. When speech is absent, the measured spectral energy is attributed to noise, and the noise estimate is updated based on the measured spectral energy. Therefore, it is important to distinguish between periods of speech and absence of speech in order to obtain an accurate noise energy estimate for computation of the SNR.
- An exemplary technique for speech detection uses a voice metric calculator to perform the noise update decision. A voice metric is a measurement of the overall voice-like characteristics of the channel energy. First, raw SNR estimates are used to index a voice metric table to obtain voice metric values for each channel. The individual channel voice metric values are summed to create an energy parameter, which is compared with a background noise update threshold. If the voice metric sum meets or exceeds the threshold, then the signal is said to contain speech. If the voice metric sum does not meet the threshold, the input frame is deemed to be noise, and a background noise update is performed. However, for the case of a high background noise condition, a sudden background noise, or an increasing noise source, SNR measurements will be large, resulting in a high voice metric, which negates a noise estimate update.
- A refinement to the voice metric calculator technique measures the channel energy deviation. This method assumes that noise exhibits constant spectral energy over time, while speech exhibits variable spectral energy over time. Thus, the channel energy is integrated over time, and speech is detected if there is substantial channel energy deviation, while noise is detected if there is little channel energy deviation. A speech detector which measures channel energy deviation will detect a sudden increase in the level of noise. However, the channel energy deviation method provides an inaccurate result when the input speech signal is of constant energy. Furthermore, for the case of an increasing noise source, changes in the input energy will cause the energy deviation to be large, negating a noise estimate update even though an update is necessary.
- In addition to an accurate speech detector, the noise suppression system must appropriately adjust channel gains. Channel gains should be adjusted so that noise suppression is achieved without sacrificing the voice quality. One method of channel gain adjustment computes the gain as a function of the total noise estimate and the SNR of the speech signal. In general, an increase in the total noise estimate results in a lower gain factor for a given SNR. A lower gain factor is indicative of a greater attenuation factor. This technique imposes a minimum gain value to prevent excess attenuation of the channel gain when the total noise estimate is very high. By using a hard clamped minimum gain value, a tradeoff between noise suppression and voice quality is introduced. When the clamp is relatively low, noise suppression is improved but voice quality is degraded. When the clamp is relatively high, noise suppression is degraded but the voice quality is improved.
- US Patent No. 4, 811, 404 "Motorola Inc.", discloses a method and apparatus for suppressing background noise in high background noise environments. This US Patent includes the addition of a signal to noise ratio (SNR) threshold mechanism to reduce background noise flutter by off-setting the gain rise of a gain table until a certain SNR threshold is reached.
- In order to provide an improved noise suppression system, the limitations of the current techniques for speech detection and channel gain computation need to be addressed. These problems and deficiencies are solved by the present invention in the manner described below.
- The present invention as set out in the appended claims is a noise suppression system and method for use in speech processing systems. An objective of the present invention is to provide a speech detector which determines the presence of speech in an input signal. A reliable speech detector is needed for an accurate determination of the signal-to-noise ratio (SNR) of speech. When speech is determined to be absent, the input signal is assumed to be entirely a noise signal, and the noise energy may be measured. The noise energy is then used for determination of the SNR. Another objective of the present invention is to provide an improved gain determination element for realization of noise suppression.
- In accordance with the present invention, the noise suppression system comprises a speech detector which determines if speech is present in a frame of the input signal. The speech decision may be based on the SNR measure of speech in an input signal. A SNR estimator estimates the SNR based on the signal energy estimate generated by an energy estimator and the noise energy estimate generated by a noise energy estimator. The speech decision may also be based on the encoding rate of the input signal. In a variable rate communication system, each input frame is assigned an encoding rate selected from a predetermined set of rates based on the content of the input frame. Generally, the rate is dependent on the level of speech activity, so that a frame containing speech would be assigned -a high rate, whereas a frame not containing speech would be assigned a low rate. Further, the speech decision may be based on one or more mode measures which are descriptive of the characteristics of the input signal. If it is determined that speech is not present in the input frame, then the noise energy estimator updates the noise energy estimate.
- A channel gain estimator determines the gain for the frame of input signal. If speech is not present in the frame, then the gain is set to be a predetermined minimum. Otherwise, the gain is determined based on the frequency content of the frame. In a preferred embodiment, a gain factor is determined for each of a set of predefined frequency channels. For each channel, the gain is determined in accordance with the SNR of the speech in the channel. For each channel, the gain is defined using a function that is suitable for the characteristics of the frequency band within which the channel is located. Typically, for a predefined frequency band, the gain is set to increase linearly with increasing SNR. Additionally, the minimum gain for each frequency band may be adjustable based on the environmental characteristics. For example, a user-selectable minimum gain may be implemented. The channel SNRs are based on channel energy estimates generated by an energy estimator and channel noise energy estimates generated by a noise energy estimator. The gain factors are used to adjust the gain of the signal in the different channels, and the gain adjusted channels are combined to produce the noise suppressed output signal.
- The features, objects, and advantages of the present invention will become more apparent from the detailed description set forth below when taken in conjunction with the drawings in which like reference characters identify correspondingly throughout and wherein:
- FIG. 1 is a block diagram of a communications system in which a noise suppressor is utilized;
- FIG. 2 is a block diagram illustrating a noise suppressor in accordance with the present invention;
- FIG. 3 is a graph of gain factors based on frequency, for realization of noise suppression in accordance with the present invention; and
- FIG. 4 is a flow chart illustrating an exemplary embodiment of the processing steps involved in noise suppression as implemented by the processing elements of FIG. 2.
- In speech communication systems, noise suppressors are commonly used to suppress undesirable environmental background noise. Most noise suppressors operate by estimating the background noise characteristics of the input data signal in one or more frequency bands and subtracting an average of the estimate(s) from the input signal. The estimate of the average background noise is updated during periods of the absence of speech. Noise suppressors require an accurate determination of the background noise level for proper operation. In addition, the level of noise suppression must be properly adjusted based on the speech and noise characteristics of the input signal. These requirements are addressed by the noise suppression system of the present invention.
- An exemplary
speech processing system 100 in which the present invention may be embodied is illustrated in FIG. 1.System 100 comprisesmicrophone 102, A/D converter 104,speech processor 106,transmitter 110, andantenna 112.Microphone 102 may be located in a cellular telephone together with the other elements illustrated in FIG. 1. Alternatively,microphone 102 may be the hands-free microphone of the vehicle speakerphone option to a cellular communication system. The vehicle speakerphone assembly is sometimes referred to as a carkit. Wheremicrophone 102 is part of a carkit, the noise suppression function is particularly important. Because the hands-free microphone is generally positioned at some distance from the user, the received acoustic signal tends to have a poor speech SNR due to road and wind noise conditions. - Referring still to FIG. 1, the input audio signal, comprising speech and/or background noise, is received by
microphone 102. The input audio signal is transformed bymicrophone 102 into an electro-acoustic signal represented by the term s(t). The electro-acoustic signal may be converted from an analog signal to pulse code modulated (PCM) samples by Analog-to-Digital converter 104. In an exemplary embodiment, PCM samples are output by A/D converter 104 at 64 kbps and are represented by signal s(n) as shown in FIG. 1. Digital signal s(n) is received byspeech processor 106, which comprises, among other elements,noise suppressor 108.Noise suppressor 108 suppresses noise in signal s(n) in accordance with the present invention. In a carkit application,noise suppressor 108 determines the level of background environmental noise and adjusts the gain of the signal to mitigate the effects of such environmental noise. In addition tonoise suppressor 108,speech processor 106 generally comprises a voice coder, or a vocoder (not shown), which compresses speech by extracting parameters that relate to a model of human speech generation.Speech processor 106 may also comprise an echo canceller (not shown), which eliminates acoustic echo resulting from the feedback between a speaker (not shown) andmicrophone 102. - Following processing by
speech processor 106, the signal is provided totransmitter 110, which performs modulation in accordance with a predetermined format such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), or Frequency Division Multiple Access (FDMA). In the exemplary embodiment,transmitter 110 modulates the signal in accordance with a CDMA modulation format as described in U.S. Pat. No. 4,901,307, entitled "SPREAD SPECTRUM MULTIPLE ACCESS COMMUNICATION SYSTEM USING SATELLITE OR TERRESTRIAL REPEATERS," which is assigned to the assignee of the present invention.Transmitter 110 then upconverts and amplifies the modulated signal, and the modulated signal is transmitted throughantenna 112. - It should be recognized that
noise suppressor 108 may be embodied in speech processing systems that are not identical tosystem 100 of FIG. 1. For example,noise suppressor 108 may be utilized within an electronic mail application having a voice mail option. For such an application,transmitter 110 andantenna 112 of FIG. 1 will not be necessary. Instead, the noise suppressed signal will be formatted byspeech processor 106 for transmission through the electronic mail network. - An exemplary embodiment of
noise suppressor 108 is illustrated in FIG. 2. The input audio signal is received bypreprocessor 202, as shown in FIG. 2.Preprocessor 202 prepares the input signal for noise suppression by performing preemphasis and frame generation. Preemphasis redistributes the power spectral density of the speech signal by emphasizing the high frequency speech components of the signal. Essentially performing a high pass filtering function, preemphasis emphasizes the important speech components to enhance the SNR of these components in the frequency domain.Preprocessor 202 may also generate frames from the samples of the input signal. In a preferred embodiment, 10 ms frames of 80 samples/frame are generated. The frames may have overlapped samples for- better processing accuracy. The frames may be generated by windowing and zero padding of the samples of the input signal. The preprocessed signal is presented to transformelement 204. In a preferred embodiment, transformelement 204 generates a 128 point Fast Fourier Transform (FFT) for each frame of input signal. It should be understood, however, that alternative schemes may be used to analyze the frequency components of the input signal. - The transformed components are provided to channel
energy estimator 206a, which generates an energy estimate for each of N channels of the transformed signal. For each channel, one technique for updating the channel energy estimates the update to be the current channel energy smoothed over channel energies of previous frames as follows:
where the updated estimate, Eu(t), is defined as a function of the current channel energy, Ech, and the previous estimated channel noise energy, Eu(t-1). An exemplary embodiment sets α = 0.55. - A preferred embodiment determines an energy estimate for a low frequency channel and an energy estimate for a high frequency channel, so that N=2. The low frequency channel corresponds to frequency range from 250 to 2250 Hz, while the high frequency channel corresponds to frequency range from 2250 to 3500 Hz. The current channel energy of the low frequency channel may be determined by summing the energy of the FFT points corresponding to 250-2250 Hz, and the current channel energy of the high frequency channel may be determined by summing the energy of the FFT points corresponding to 2250-3500 Hz.
- The energy estimates are provided to
speech detector 208, which determines whether or not speech is present in the received audio signal.SNR estimator 210a ofspeech detector 208 receives the energy estimates.SNR estimator 210a determines the signal-to-noise ratio (SNR) of the speech in each of the N channels based on the channel energy estimates and the channel noise energy estimates. The channel noise energy estimates are provided bynoise energy estimator 214a, and generally correspond to the estimated noise energy smoothed over the previous frames which do not contain speech. -
Speech detector 208 also comprisesrate decision element 212, which selects the data rate of the input signal from a predetermined set of data rates. In certain communication systems, data is encoded so that the data rate may be varied from one frame to another. This is known as a variable rate communication system. The voice coder which encodes data based on a variable rate scheme is typically called a variable rate vocoder. An exemplary embodiment of a variable rate vocoder is described in U.S. Pat. No. 5,414,796, entitled "VARIABLE RATE VOCODER," assigned to the assignee of the present invention. The use of a variable rate communications channel eliminates unnecessary transmissions when there is no useful speech to be transmitted. Algorithms are utilized within the vocoder for generating a varying number of information bits in each frame in accordance with variations in speech activity. For example, a vocoder with a set of four rates may produce 20 millisecond data frames containing 16, 40, 80, or 171 information bits, depending on the activity of the speaker. It is desired to transmit each data frame in a fixed amount of time by varying the transmission rate of communications. - Because the rate of a frame is dependent on the speech activity during a time frame, determining the rate will provide information on whether speech is present or not. In a system utilizing variable rates, a determination that a frame should be encoded at the highest rate generally indicates the presence of speech, while a determination that a frame should be encoded at the lowest rate generally indicates the absence of speech. Intermediate rates typically indicate transitions between the presence and the absence of speech.
-
Rate decision element 212 may implement any of a number of rate decision algorithms. One such rate decision algorithm is disclosed in copending U.S. patent application Serial No. 08/286,842, entitled "METHOD AND APPARATUS FOR PERFORMING REDUCED RATE VARIABLE RATE VOCODING," assigned to the assignee of the present invention. This technique provides a set of rate decision criteria referred to as mode measures. A first mode measure is the target matching signal to noise ratio (TMSNR) from the previous encoding frame, which provides information on how well the encoding model is performing by comparing a synthesized speech signal with the input speech signal. A second mode measure is the normalized autocorrelation function (NACF), which measures periodicity in the speech frame. A third mode measure is the zero crossings (ZC) parameter, which measures high frequency content in an input speech frame. A fourth measure, the prediction gain differential (PGD), determines if the encoder is maintaining its prediction efficiency. A fifth measure is the energy differential (ED), which compares the energy in the current frame to an average frame energy. Using these mode measures, a rate determination logic selects an encoding rate for the frame of input. - It should be understood that although
rate decision element 212 is shown in FIG. 2 as an included element ofnoise suppressor 108, the rate information may instead be provided tonoise suppressor 108 by another component of speech processor 106 (FIG. 1). For example,speech processor 106 may comprise a variable rate vocoder (not shown) which determines the encoding rate for each frame of input signal. Instead of havingnoise suppressor 108 independently perform rate determination, the rate information may be provided tonoise suppressor 108 by the variable rate vocoder. - It should also be understood that instead of using the rate decision to determine the presence of speech,
speech detector 208 may use a subset of the mode measures that contribute to the rate decision. For instance,rate decision element 212 may be substituted by a NACF element (not shown), which, as explained earlier, measures periodicity in the speech frame. The NACF is evaluated in accordance with the relationship below:
where N refers to the numbers of samples of the speech frame, t1 and t2 refer to the boundaries within the T samples for which the NACF is evaluated. The NACF is evaluated based on the formant residual signal, e(n). Formant frequencies are the resonance frequencies of speech. A short term filter is used to filter the speech signal to obtain the formant frequencies. The residual signal obtained after filtering by the short term filter is the formant residual signal, and contains the long term speech information, such as the pitch, of the signal. - The NACF mode measure is suitable for determining the presence of speech because the periodicity of a signal containing voiced speech is different from a signal which does not contain voiced speech. A voiced speech signal tends to be characterized by periodic components.When voiced speech is not present, the signal generally will not have periodic components. Thus, the NACF measure is a good indicator which may be used by
speech detector 208. -
Speech detector 208 may use measures such as the NACF instead of the rate decision in situations where it is not practicable to generate the rate decision. For example, if the rate decision is not available from the variable rate vocoder, andnoise processor 108 does not have the processing power to generate its own rate decision, then mode measures like the NACF offer a desirable alternative. This may be the case in a carkit application where processing power is generally limited. - Additionally, it should be understood that
speech detector 208 may make a determination regarding the presence of speech based on the rate decision, the mode measure(s), or the SNR estimate alone. Although additional measures should improve the accuracy of the determination, any one of the measures alone may provide an adequate result. - The rate decision (or the mode measure(s)) and the SNR estimate generated by
SNR estimator 210a are provided tospeech decision element 216.Speech decision element 216 generates a decision on whether or not speech is present in the input signal based on its inputs. The decision on the presence of speech will determine if a noise energy estimate update should be performed. The noise energy estimate is used bySNR estimator 210a to determine the SNR of the speech in the input signal. The SNR will in turn will be used to compute the level of attenuation of the input signal for noise suppression. If it is determined that speech is present, thenspeech decision element 216 opensswitch 218a, preventingnoise energy estimator 214a from updating the noise energy estimate. If it is determined that speech is not present, then the input signal is assumed to be noise, andspeech decision element 216 closes switch 218a, causingnoise energy estimator 214a to update the noise estimate. Although shown in FIG. 2 asswitch 218a, it should be understood that an enable signal provided byspeech decision element 216 tonoise energy estimator 214a may perform the same function. - In a preferred embodiment in which two channel SNRs are evaluated,
speech decision element 216 generates the noise update decision based on the procedure below: The channel SNR estimates provided bySNR estimator 210a are denoted by chsnr1 and chsnr2. The rate of the input signal, provided byrate decision element 212, is denoted by rate. A counter, ratecount, keeps track of the number of frames based on certain conditions as described below. -
Speech decision element 216 determines that speech is not present, and that the noise estimate should be updated, if the rate is the minimum rate of the variable rates, either chsnr1 is greater than threshold T1 or chsnr2 is greater than threshold T2, and ratecount is greater than threshold T3. If the rate is minimum, and either chsnr1 is greater than T1 or chsnr2 is greater than T2, but ratecount is less than T3, then the ratecount is increased by one but no noise estimate update is performed. The counter, ratecount, detects the case of a sudden increased level of noise or an increasing noise source by counting the number of frames having minimum rate but also having high energy in at least one of the channels. The counter, which provides an indicator that the high SNR signal contains no speech, is set to count until speech is detected in the signal. A preferred embodiment sets T1 = T2 = 5 dB, and T2 = 100 frames where 10 ms frames are evaluated. - If the rate is minimum, chsnr1 is less than T1, and chsnr2 is less than T2, then
speech decision element 216 will determine that speech is not present and that a noise estimate update should be performed. In addition, ratecount is reset to zero. - If the rate is not minimum, then
speech decision element 216 will determine that the frame contains speech, and no noise estimate update is performed, but ratecount is reset to zero. - Instead of using the rate measure to determine the presence of speech, recall that mode measures such as a NACF measure may be utilized instead.
Speech decision element 216 may make use of the NACF measure to determine the presence of speech; and thus the noise update decision, in accordance with the procedure below: where pitchPresent is defined as follows: Again, channel SNR estimates provided bySNR estimator 210a are denoted by chsnr1 and chsnr2. A NACF element (not shown) generates a measure indicative of the presence of pitch, pitchPresent, as defined above. A counter, pitchCount, keeps track of the number of frames based on certain conditions as described below. - The measure pitchPresent determines that pitch is present if NACF is above threshold TT1. If NACF falls within a mid range (TT2 ≤ NACF ≤ TT1) for a number of frames greater than threshold TT3, then pitch is also determined to be present. A counter, NACFcount, keeps track of the number of frames for which TT2 ≤ NACF ≤ TT1. In a preferred embodiment, TT1 = 0.6, TT2 = 0.4, and TT3 = 8 frames where 10 ms frames are evaluated.
-
Speech decision element 216 determines that speech is not present, and that the noise estimate should be updated, if the pitchPresent measure indicates that pitch is not present (pitchPresent = FALSE), either chsnr1 is greater than threshold TH1 or chsnr2 is greater than threshold TH2, and pitchCount is greater than threshold TH3. If pitchPresent = FALSE, and either chsnr1 is greater than TH1 or chsnr2 is greater than TH2, but pitchCount is less than TH3, then pitchCount is increased by one but no noise estimate update is performed. The counter, pitchCount, is used to detect the case of a sudden increased level of noise or an increasing noise source. A preferred embodiment sets T1 = T2 = 5 dB, and T2 = 100 frames where 10 ms frames are evaluated. - If pitchPresent indicates that pitch is not present, and chsnr1 is less than TH1 and chsnr2 is less than TH2, then
speech decision element 216 will determine that speech is not present and that a noise estimate update should be performed. In addition, pitchCount is reset to zero. - If pitchPresent indicates that pitch is present (pitchPresent = TRUE), then
speech decision element 216 will determine that the frame contains speech, and no noise estimate update is performed. However, pitchCount is reset to zero. - Upon determination that speech is not present,
switch 218a is closed, causingnoise energy estimator 214a to update the noise estimate.Noise energy estimator 214a generally generates a noise energy estimate for each of the N channels of the input signal. Since speech is not present, the energy is presumed to be wholly contributed by noise. For each channel, the noise energy update is estimated to be the current channel energy smoothed over channel energies of previous frames which do not contain speech. For example, the updated estimate may be obtained based on the relationship below:
where the updated estimate, En(t), is defined as a function of the current channel energy, Ech, and the previous estimated channel noise energy, En(t-1). An exemplary embodiment sets β = 0.1. The updated channel noise energy estimates are presented toSNR estimator 210a. These channel noise energy estimates will be used to obtain channel SNR estimate updates for the next frame of input signal. - The determination regarding the presence of speech is also provided to channel
gain estimator 220.Channel gain estimator 220 determines the gain, and thus the level of noise suppression, for the frame of input signal. Ifspeech decision element 216 has determined that speech is not present, then the gain for the frame is set at a predetermined minimum gain level. Otherwise, the gain is determined as a function of frequency. In a preferred embodiment, the gain is computed based on the graph shown in FIG. 3. Although shown in graphical form in FIG. 3, it should be understood that the function illustrated in FIG. 3 may be implemented as a look-up table inchannel gain estimator 220. - Referring to FIG. 3, it can be seen that a preferred embodiment of the present invention defines a separate gain curve for each of L frequency bands. In FIG. 3, three bands (L = 3) are represented, although L may be any number greater than or equal to one. Thus, the gain factor for a channel in the low band may be determined using the low band curve, the gain factor for a channel in the mid band may be determined using the mid band curve, and the gain factor for a channel in the high band may be determined using the high band curve.
- Although noise suppression may be performed by utilizing just one gain curve for the input signal (L = 1), the use of multiple bands has been found to provide less voice quality degradation. In the case of environmental noise, such as road and wind noise, the energy of the noise signal is greater at the lower frequencies, and the energy generally decreases with increasing frequency.
-
- The preferred embodiment assigns the low band as 125 - 375 Hz, the mid band as 375 - 2625 Hz, and the high band as 2625 - 4000 Hz. The slopes and the y intercepts are experimentally determined. The preferred embodiment uses the same slope, 0.39, for each of the three bands, although a different slope may be used for each frequency band. Also, lowBandYintercept is set at -17 dB, midBandYintercept is set at -13 dB, and highBandYintercept is set at -13 dB.
- An optional feature would provide the user of the device comprising the noise suppressor to select the desired y-intercepts. Thus, more noise suppression (a lower y-intercept) may be chosen at the expense of some voice degradation. Alternatively, the y-intercepts may be variable as a function of some measure determined by
noise suppressor 108. For example, more noise suppression (a lower y-intercept) may be desired when an excessive noise energy is detected for a predetermined period of time. Alternatively, less noise suppression (a high y-intercept) may be desired when a condition such as babble is detected. During a babble condition, background speakers are present, and less noise suppression may be warranted to prevent cut out of the main speaker. Another optional feature would provide for selectable slopes of the gain curves. Further, it should be understood that a curve other than the lines described by equations (4)-(6) may be found to be more suitable for determining the gain factor under certain circumstances. - For each frame containing speech, a gain factor is determined for each of M frequency channels of the input signal, where M is the predetermined number of channels to be evaluated. A preferred embodiment evaluates sixteen channels (M=16). Referring again to FIG. 3, the gain factors for the channels having frequency components in the range of the low band are determined using the low band curve. The gain factors for the channels having frequency components in the range of the mid band are determined using the mid band curve. The gain factors for the channels having frequency components in the range of the high band are determined using the high band curve.
- For each channel evaluated, the channel SNR is used to derive the gain factor based on the appropriate curve. The channel SNRs are shown, in FIG. 2, to be evaluated by
channel energy estimator 206b,noise energy estimator 214b, andSNR estimator 210b. For each frame of input signal,channel energy estimator 206b generates energy estimates for each of M channels of the transformed input signal, and provides the energy estimates toSNR estimator 210b. The channel energy estimates may be updated using the relationship of Equation (1) above. If it is determined byspeech decision element 216 that no speech is present in the input signal, then switch 218b is closed, andnoise energy estimator 214b updates the estimates of the channel noise energy. For each of the M channels, the updated noise energy estimate is based on the channel energy estimate determined bychannel energy estimator 206b. The updated estimate may be evaluated using the relationship of Equation (3) above. The channel noise estimates are provided toSNR estimator 210b. Thus,SNR estimator 210b determines channel SNR estimates for each frame of speech based on the channel energy estimates for the particular frame of speech and the channel noise energy estimates provided bynoise energy estimator 214b. - An artisan skilled in the art would recognize that
channel energy estimator 206a,noise energy estimator 214a,switch 218a, andSNR estimator 210a perform functions similar tochannel energy estimator 206b,noise energy estimator 214b,switch 218b, andSNR estimator 210b, respectively. Thus, although shown as separate processing elements in FIG. 2,channel energy estimators noise energy estimators SNR estimators speech decision element 216, and provides the M SNR estimates tochannel gain estimator 220. - The channel gain factors are provided by
channel gain estimator 220 to gainadjuster 224.Gain adjuster 224 also receives the FFT transformed input signal fromtransform element 204. The gain of the transformed signal is appropriately adjusted according to the channel gain factors. For example, in the embodiment described above wherein M = 16, the transformed (FFT) points belonging to the particular one of the sixteen channels are adjusted based on the appropriate channel gain factor. - The gain adjusted signal generated by
gain adjuster 224 is then provided toinverse transform element 226, which in a preferred embodiment generates the Inverse Fast Fourier Transform (IFFT) of the signal. The inverse transformed signal is provided to postprocessing element 228. If the frames of input had been formed with overlapped samples, then postprocessing element 228 adjusts the output signal for the overlap.Post processing element 228 also performs deemphasis if the signal had undergone preemphasis. Deemphasis attenuates the frequency components that were emphasized during preemphasis. The preemphasis/deemphasis process effectively contributes to noise suppression by reducing the noise components lying outside of the range of the processed frequency components. - It should be understood that the various processing blocks of the noise suppressor shown in FIG. 2 may be configured in a digital signal processor (DSP) or an application specific integrated circuit (ASIC). The description of the functionality of the present invention would enable one of ordinary skill to implement the present invention in a DSP or an ASIC without undue experimentation.
- Referring now to FIG. 4, a flow chart is shown illustrating some of the steps involved in the processing as discussed with reference to FIGS. 2 and 3. Although shown as consecutive steps, one skilled in the art would recognize that ordering of some of the steps are interchangeable.
- The process begins at
step 402. Atstep 404, transformelement 204 transforms the input audio signal into a transformed signal, generally a FFT signal. Atstep 406,SNR estimator 210b determines the speech SNR for M channels of the input signal based on the channel energy estimates provided bychannel energy estimator 206b and the channel noise energy estimates provided bynoise energy estimator 214b. Atstep 408,channel gain estimator 220 determines gain factors for the M channels of the input signal based on the frequency of the channels.Channel gain estimator 220 sets the gain at a minimum level if speech has been found to be absent in the frame of input signal. Otherwise, a gain factor is determined, for each of the M channels, based on a predetermined function. For example, referring to FIG. 3, a function defined by line equations having fixed slopes and y-intercepts, wherein each line equation defines the gain for a predetermined frequency band, may be used. Atstep 410,gain adjuster 224 adjusts the gain of the M channels of the transformed signal using the M gain factors. Atstep 412,inverse transform element 226 inverse transforms the gain adjusted transformed signal, producing the noise suppressed audio signal. - At
step 414,SNR estimator 210a determines the speech SNR for N channels of the input signal based on the channel energy estimates provided bychannel energy estimator 206a and the channel noise energy estimates provided bynoise energy estimator 214a. Atstep 416,rate decision element 212 determines the encoding rate for the input signal through analysis of the input signal. Alternatively, one or more mode measures, such as the NACF, may be determined. Atstep 418,speech decision element 216 determines if speech is present in the input signal based on the SNR provided bySNR estimator 210a, the rate provided byrate decision element 212, and/or the mode measure(s). If it is determined, at decision black 420, that speech is not present, then the input signal is assumed to be entirely noise, and a noise estimate update is performed bynoise energy estimator 214a atstep 422.Noise energy estimator 214a updates the noise estimate based on the channel energy determined bychannel energy estimator 206a. Whether or not speech is detected, the procedure continues to process the next frame of the input signal. - The previous description of the preferred embodiments is provided to enable any person skilled in the art to make or use the present invention. The scope of the invention is defined by the appended claims.
Claims (32)
- A noise suppressor (108) for suppressing the background noise of an audio signal, comprising:a signal to noise ratio (SNR) estimator (210b) for generating channel SNR estimates for a first predefined set of frequency channels of said audio signal;a gain estimator (220) for generating a gain factor for each of said frequency channels based on a corresponding one of said channel SNR estimates, wherein said gain factor is derived using a gain function which defines gain factor as an increasing function of SNR;a gain adjuster (224) for adjusting the gain level of each of said frequency channels based on said corresponding gain factor; anda speech detector (208) for determining the presence of speech in said audio signal,characterised in thatsaid speech detector (208) comprises a speech decision element (216) for determining the presence of speech in accordance with SNR estimates for a second predefined set of frequency channels of said audio signal, and in accordance with eithera) an encoding rate of a set of encoding rates for the audio signal, orb) at least one mode measure characterizing the audio signal.
- The noise suppressor (108) of claim 1 wherein the speech detector (208) further comprises a further signal to noise ratio (SNR) estimator (210a) for generating said SNR estimates for a second predefined set of frequency channels of said audio signal.
- The noise suppressor (108) of claim 2 wherein the speech detector (208) further comprises a rate decision element (212) for determining said encoding rate of a set of encoding rates for the audio signal.
- The noise suppressor (108) of claim 2 wherein the speech detector (208) further comprises a mode measurement element for determining said at least one mode measure.
- The noise suppressor (108) of any preceding claim wherein said mode measures comprise a normalized autocorrelation function (NACF) measure.
- The noise suppressor (108) of any preceding claim wherein said gain function is frequency dependent.
- The noise suppressor (108) of any preceding claim wherein said gain function is implemented as a look-up table.
- The noise suppressor (108) of any preceding claim wherein said gain function is a linear function having a slope and a y-intercept.
- The noise suppressor (108) of claim 8, wherein said y-intercept is user selectable.
- The noise suppressor (108) of claim 8, wherein said y-intercept is adjustable based on the measured characteristics of noise in said audio signal.
- The noise suppressor (108) of claim 8, wherein said slope is user selectable.
- The noise suppressor (108) of claim 8 wherein said slope is adjustable based on the measured characteristics of noise in said audio signal.
- The noise suppressor (108) of any preceding claim, further comprising:a noise energy estimator (214b) for generating an updated channel noise energy estimate for each of said frequency channels when said speech detector (208) determines that speech is not present in said audio signal, said updated channel noise energy estimates provided to said SNR estimator (210b) for generating said channel SNR estimates.
- The noise suppressor (108) of any preceding claim wherein said gain estimator (220) determines a minimum gain factor for each of said frequency channels if said speech detector determines that speech is not present.
- The noise suppressor (108) of any preceding claim wherein said gain estimator (220) comprises:means for determining a gain factor for each of said frequency channels if said means for determining the presence of speech determines that speech is present, wherein a gain function is defined for each of a set of frequency bands, and for each said frequency band, gain factor is defined to increase with increasing SNR, so that for each of said frequency channels, a channel gain factor is determined based on the gain function for the frequency band whose range contains the frequency channel.
- The noise suppressor (108) of any preceding claim further comprising:means for transforming said audio signal into a frequency representation of said audio signal; andmeans for inverse transforming said gain adjusted frequency representation to generate a noise suppressed audio signal.
- A method for suppressing the background noise of an audio signal, comprising the steps of:generating channel SNR estimates for a first predefined set of frequency channels of said audio signal;generating a gain factor for each of said frequency channels based on a corresponding one of said channel SNR estimates, wherein said gain factor is derived using a gain function which defines gain factor as an increasing function of SNR:adjusting the gain level of each of said frequency channels based on said corresponding gain factor; anddetermining the presence of speech in said audio signal,characterised in thatthe presence of speech is determined in accordance with SNR estimates for a second predefined set of frequency channels of said audio signal and in accordance with eithera) an encoding rate of a set of encoding rates for the audio signal, orb) at least one mode measure characterizing the audio signal.
- The method of claim 17 further comprising the step of generating said SNR estimates for a second predefined set of frequency channels of said audio signal.
- The method of claim 18 further comprising the step of determining said encoding rate of a set of encoding rates for the audio signal.
- The method of claim 18 further comprising the step of determining said at least one mode measure.
- The method of any of claims 17 to 20, wherein said gain function is frequency dependent.
- The method of any of claims 17 to 21 wherein said gain function is implemented as a look-up table.
- The method of any of claims 17 to 22 wherein each of said gain functions is a linear function having a slope an a y-intercept.
- The method claim 23, wherein said y-intercept is user selectable.
- The method of claim 23, wherein said y-intercept is adjustable based on the measured characteristics of noise in said audio signal.
- The method of claim 23, wherein said slope is user selectable.
- The method of claim 23, wherein said slope is adjustable based on the measured characteristics of noise in said audio signal.
- The method of any of claims 17 to 27 further comprising the step of:generating an updated channel noise energy estimate for each of said frequency channels when said step of determining the presence of speech determines that speech is absent in said audio signal, said updated channel noise energy estimates to be used for generating said channel SNR estimates.
- The method of any of claims 17 to 28 further comprising the steps of:transforming said audio signal into a frequency representation of said audio signal; andinverse transforming said gain adjusted frequency representation to generate a noise suppressed audio signal.
- The method of any of claims 17 to 29 wherein said step of generating a gain factor comprising the step of:determining a gain factor for each of said frequency channels if speech is determined to be present in said audio signal, wherein a gain function is defined for each of a set of frequency bands, and for each said frequency band, gain is defined to increase with increasing SNR, so that for each of said frequency channels, a channel gain factor is determined based on the gain function for the frequency band whose range contains the frequency channel.
- The method of any of claims 17 to 30 wherein said mode measures comprise a normalized autocorrelation function (NACF) measure.
- The method of any of claims 17 to 31 further comprising the step of:determining a minimum gain factor for each of said frequency channels if speech is determined to be absent in said audio signal.
Applications Claiming Priority (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
1997-05-09 | |||
US08/921,492 US6122384A (en) | 1997-09-02 | 1997-09-02 | Noise suppression system and method |
PCT/US1997/017656 WO1999012155A1 (en) | 1997-09-30 | 1997-09-30 | Channel gain modification system and method for noise reduction in voice communication |
Publications (2)
Publication Number | Publication Date |
---|---|
EP1010169A1 EP1010169A1 (en) | 2000-06-21 |
EP1010169B1 true EP1010169B1 (en) | 2006-06-21 |
Family
ID=56289795
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
EP97945400A Expired - Lifetime EP1010169B1 (en) | 1997-09-02 | 1997-09-30 | Channel gain modification system and method for noise reduction in voice communication |
Country Status (3)
Country | Link |
---|---|
EP (1) | EP1010169B1 (en) |
JP (1) | JP4194749B2 (en) |
DE (1) | DE69736198T2 (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2001318694A (en) * | 2000-05-10 | 2001-11-16 | Toshiba Corp | Device and method for signal processing and recording medium |
JP4580508B2 (en) * | 2000-05-31 | 2010-11-17 | 株式会社東芝 | Signal processing apparatus and communication apparatus |
JP5131149B2 (en) * | 2008-10-24 | 2013-01-30 | ヤマハ株式会社 | Noise suppression device and noise suppression method |
-
1997
- 1997-09-30 EP EP97945400A patent/EP1010169B1/en not_active Expired - Lifetime
- 1997-09-30 JP JP2000509079A patent/JP4194749B2/en not_active Expired - Fee Related
- 1997-09-30 DE DE69736198T patent/DE69736198T2/en not_active Expired - Lifetime
Also Published As
Publication number | Publication date |
---|---|
EP1010169A1 (en) | 2000-06-21 |
DE69736198D1 (en) | 2006-08-03 |
JP4194749B2 (en) | 2008-12-10 |
DE69736198T2 (en) | 2007-05-03 |
JP2003526109A (en) | 2003-09-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US6122384A (en) | Noise suppression system and method | |
US6233549B1 (en) | Low frequency spectral enhancement system and method | |
US9646621B2 (en) | Voice detector and a method for suppressing sub-bands in a voice detector | |
US7555075B2 (en) | Adjustable noise suppression system | |
EP0790599B1 (en) | A noise suppressor and method for suppressing background noise in noisy speech, and a mobile station | |
US9761246B2 (en) | Method and apparatus for detecting a voice activity in an input audio signal | |
US7058572B1 (en) | Reducing acoustic noise in wireless and landline based telephony | |
EP0707763B1 (en) | Reduction of background noise for speech enhancement | |
US7171246B2 (en) | Noise suppression | |
US5544250A (en) | Noise suppression system and method therefor | |
US7873114B2 (en) | Method and apparatus for quickly detecting a presence of abrupt noise and updating a noise estimate | |
US7912729B2 (en) | High-frequency bandwidth extension in the time domain | |
US20050108004A1 (en) | Voice activity detector based on spectral flatness of input signal | |
US20070232257A1 (en) | Noise suppressor | |
US6510224B1 (en) | Enhancement of near-end voice signals in an echo suppression system | |
US20050267741A1 (en) | System and method for enhanced artificial bandwidth expansion | |
WO1999012155A1 (en) | Channel gain modification system and method for noise reduction in voice communication | |
EP1010169B1 (en) | Channel gain modification system and method for noise reduction in voice communication | |
US20030065509A1 (en) | Method for improving noise reduction in speech transmission in communication systems | |
JP2003517761A (en) | Method and apparatus for suppressing acoustic background noise in a communication system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
17P | Request for examination filed |
Effective date: 20000306 |
|
AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): DE FR GB |
|
RAP1 | Party data changed (applicant data changed or rights of an application transferred) |
Owner name: QUALCOMM INCORPORATED |
|
17Q | First examination report despatched |
Effective date: 20030321 |
|
GRAP | Despatch of communication of intention to grant a patent |
Free format text: ORIGINAL CODE: EPIDOSNIGR1 |
|
RIC1 | Information provided on ipc code assigned before grant |
Ipc: 7G 10L 11/02 A |
|
GRAS | Grant fee paid |
Free format text: ORIGINAL CODE: EPIDOSNIGR3 |
|
GRAA | (expected) grant |
Free format text: ORIGINAL CODE: 0009210 |
|
AK | Designated contracting states |
Kind code of ref document: B1 Designated state(s): DE FR GB |
|
REG | Reference to a national code |
Ref country code: GB Ref legal event code: FG4D |
|
REF | Corresponds to: |
Ref document number: 69736198 Country of ref document: DE Date of ref document: 20060803 Kind code of ref document: P |
|
ET | Fr: translation filed | ||
REG | Reference to a national code |
Ref country code: HK Ref legal event code: GR Ref document number: 1034136 Country of ref document: HK |
|
PLBE | No opposition filed within time limit |
Free format text: ORIGINAL CODE: 0009261 |
|
STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: NO OPPOSITION FILED WITHIN TIME LIMIT |
|
26N | No opposition filed |
Effective date: 20070322 |
|
PGFP | Annual fee paid to national office [announced via postgrant information from national office to epo] |
Ref country code: FR Payment date: 20100920 Year of fee payment: 14 |
|
PGFP | Annual fee paid to national office [announced via postgrant information from national office to epo] |
Ref country code: GB Payment date: 20100809 Year of fee payment: 14 |
|
PGFP | Annual fee paid to national office [announced via postgrant information from national office to epo] |
Ref country code: DE Payment date: 20100930 Year of fee payment: 14 |
|
GBPC | Gb: european patent ceased through non-payment of renewal fee |
Effective date: 20110930 |
|
REG | Reference to a national code |
Ref country code: FR Ref legal event code: ST Effective date: 20120531 |
|
REG | Reference to a national code |
Ref country code: DE Ref legal event code: R119 Ref document number: 69736198 Country of ref document: DE Effective date: 20120403 |
|
PG25 | Lapsed in a contracting state [announced via postgrant information from national office to epo] |
Ref country code: DE Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES Effective date: 20120403 |
|
PG25 | Lapsed in a contracting state [announced via postgrant information from national office to epo] |
Ref country code: GB Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES Effective date: 20110930 Ref country code: FR Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES Effective date: 20110930 |