CN1121684C - System for adaptively filtering audio signals to enhance speech intelligibility in noisy environmental conditions - Google Patents

Abstract

A method and system are provided for adaptively reducing noise in frames of digitized audio signals that include both speech and background noise. Frames of digitized audio signals are passed through an adjustable, high-pass filter circuit to filter a portion of background noise located in a low frequency range of the digitized signal. The filter circuit is adjusted by a filter control circuit adapted for a current frame to exhibit a selected frequency response curve. The filter control circuit includes a speech detector for detecting the presence or absence of speech in the frames of digitized audio signals. The filter circuit is adjusted when no speech is detected in the current frame. In a first preferred embodiment, the filter control circuit controls the filter circuit by calculating a noise estimate corresponding to the background noise, and adjusting the filter circuit based on the noise estimate. As the noise estimates increase, the filter circuit is adjusted to extract increasing amounts of energy falling in low frequency ranges of speech. In a second preferred embodiment, the filter circuit is adjusted as a function of a noise profile estimate. A noise profile estimate for a current frame is determined as a function of speech detection and is compared to a reference noise profile. Based on this comparison, the filter circuit is adaptively adjusted.

Description

Method and apparatus for selectively changing a frame of digital signal

Technical field

The invention relates to noise reduction systems, and more particularly to an adaptive speech intelligibility enhancement system for portable digital radiotelephones and a method and apparatus for selectively altering a frame of digital signals.

Background technique

The cellular telephone industry has made remarkable strides in commercial operations in the United States and other parts of the world. In major metropolitan areas, the demand for cellular service is exceeding the capacity of existing systems. Assuming this trend continues, cellular radio communications will reach even the smallest rural markets. Therefore, cellular capacity must be increased while maintaining a high quality of service at a reasonable cost. A major step towards increasing capacity is the conversion of cellular systems from analog to digital transmission. This transition is also important because first-generation Personal Communications Networks (PCNs) will likely be provided by cellular carriers using next-generation digital cellular Low-cost, pocket-sized cordless phone for making and receiving calls in the office, on the street, in the car, and more.

Digital communication systems utilize powerful digital signal processing techniques. Digital signal processing generally refers to the mathematical or other manipulation of digitized signals. For example, after converting (digitizing) an analog signal to digital form, the digital signal may be filtered, amplified, and attenuated using simple mathematical routines in a digital signal processor (DSP). Digital signal processors are typically fabricated as high-speed integrated circuits such that data processing operations are performed substantially in real time. Digital signal processors can also be used to reduce the bit rate of digitized speech. The result is a reduction in the spectrum occupancy of the transmitted radio signal and an increase in system capacity. For example, if 14-bit linear pulse code modulation (PCM) is used to digitize the voice signal and sampled at a sampling rate of 8KHZ, a serial bit rate of 112K bits/second will be generated. In addition, by mathematically exploiting the properties of redundancy and other predictable properties of human speech, vocoder techniques can be used to compress the serial bit rate of 112Kbit/s to 7.95Kbit/s to obtain a 14:1 bit Transmission rate reduction. Reduced transfer rates mean more available bandwidth.

In the United States, a popular speech compression technique adopted by TIA as the digitization standard for the second generation cellular telephone system (i.e. IS-54) is Vector Source Book Excited Linear Predictive Coding (VSELP). Unfortunately, when an audio signal comprising speech mixed with high levels of ambient noise (especially "colored noise") is encoded/compressed using VSELP, the result may include some undesired characteristics of the audio signal. For example, if a digital mobile phone is used in a noisy environment (e.g., inside a vehicle in which it is moving), both the ambient noise and the desired speech are compressed using the VSELP coding algorithm and sent to a base station where the compressed signal is decoded and reconstructed into audible speech. Undesirable audible distortions of the noise, as well as those that happen to occur in speech, are introduced when the background noise is reconstructed into analog form. This distortion is very annoying to the average listener.

Much of this distortion is caused by the environment in which the mobile phone is used. Mobile phones are typically used inside vehicles, where there is often ambient noise from the car engine and noise from surrounding traffic. Such ambient noise inside a vehicle is typically concentrated in the low frequency range, and the noise amplitude can vary due to factors such as the speed and acceleration of the vehicle and the amount of surrounding traffic. This low frequency noise also has a tendency to severely degrade speech intelligibility from people speaking in the car. This reduction in speech intelligibility due to low-frequency noise may be particularly pronounced in communication systems employing VSELP vocoders, but it may also occur in communication systems that do not include VSELP vocoders.

The effect of ambient noise on a mobile phone may also play a role in how the mobile phone is used. In particular the mobile phone can be used in hands-free mode, ie the user of the phone speaks into the mobile phone placed in the cradle. This frees the mobile phone user's hands for driving, but also increases the distance that the user's spoken words must travel before reaching the mobile phone's microphone input. This increased distance between the user and the mobile phone, combined with varying ambient noise, can cause noise to be a significant portion of the total power spectral energy of the audio signal input to the mobile phone.

Descriptions of the prior art include EP 0 645 756, EP 0 558 312, EP 0 665530, DE 4 012 349, U.S. Patent Nos. 4,811,404, 4,461,025, and 5,251,263, all of which describe methods for filtering out unwanted signal components Way.

Theoretically, a digital signal processor can be used to implement various digital signal processing algorithms to filter out the background noise of the VSELP code. However, these solutions often require significant digital signal processing overhead measured in millions of instructions per second (MIPS), which consumes valuable processing time, memory space, and power consumption. However, in portable radiotelephones, each of these signal processing resources is limited. Therefore, simply increasing the processing burden on the DSP is not an optimal solution for minimizing the background noise of VSLEP encoding as well as other types of background noise.

Summary of the invention

This invention presents an adaptive noise reduction system that cuts unwanted coding background noise while minimizing any negative impact on the quality of coded speech and any increased consumption of digital signal processing resources effect. The inventive method and system increase the intelligibility of speech in a digitized audio signal by passing frames of the digitized audio signal through a filter circuit. The filter circuit acts as an adjustable high-pass filter that rejects a portion of the digitized signal in the low frequency range and passes a portion of the digitized signal that falls in the high frequency range. Because the noise in the vehicle tends to be concentrated in the low frequency range, and only a small part of the speech intelligibility value falls in this low frequency range, the filter circuit filters out the noise in the digitized audio signal while only filtering out the unimportant part of the speech. most of the noise. This results in a relatively larger fraction of noise energy being removed than the fraction of speech energy being removed. By adaptively adjusting and selecting the frequency response curve of the filter circuit, the amount of speech filtered out is limited with minimal impact on speech intelligibility at the radio output.

The filter control circuit is used to adjust the filter circuit to exhibit different frequency response curves in the form of a noise estimate and/or spectral envelope function corresponding to Noise in an audio signal. The noise estimate and/or spectral envelope is adjusted to the digital signal on a frame-by-frame basis and in the form of a speech detection function. If no speech is detected, the noise estimate and/or spectral envelope are revised for the current frame. If speech is detected, the noise estimate and/or spectral envelope are not adjusted.

In a first embodiment, a filter circuit calculates a noise estimate for a frame of a digitized audio signal. The noise estimate corresponds to the amount of background noise in the frame of the digitized audio signal. The noise estimate increases as the relative amount of background noise in the low frequency range of speech to speech increases. As the relative amount of background noise to speech in the low frequency range of speech increases, the filter control circuit uses the noise estimate to adjust the filter circuit to filter out a greater portion of the low frequency range speech. When no background noise is present, no speech signal is filtered out. When a higher noise level is present, a larger portion of noise and speech information is extracted. Because noise tends to be concentrated in the low frequency range and only a relatively small portion of the speech intelligibility value falls within this low frequency range, as the noise estimate increases, the overall intelligibility of the audio signal is improved by increasing the portion of low frequency energy that is being filtered out can be improved.

In a second embodiment, a modified filter control circuit is used to adjust the filter circuit to exhibit different frequency response curves as a function of a noise envelope, where the noise envelope is the audio The noise envelope of the noise estimate over the selected frequency range in the signal. The filter control circuit includes a spectral analyzer that determines a noise profile estimate as a function of the detected speech. An estimate of the noise profile is determined for the current frame and compared to the reference noise profile. Based on this comparison, the filter circuit is adaptively adjusted to extract different amounts of low frequency energy from the current frame.

The adaptive clipping system according to the invention can be advantageously applied to radio communication systems in which portable/mobile radio transceivers communicate with each other and between radio transceivers and fixed telephone line subscribers via RF channels. Each transceiver includes an antenna, a receiver for converting radio signals received on an RF channel through the antenna into analog audio signals, and a transmitter. The transmitter includes a coder-decoder (codec) for digitizing the analog audio signal to be transmitted into frames of digitized speech information, the speech information including both speech and background noise. The digital signal processor processes the current frame to minimize the background noise based on the background noise estimate and speech detection in the current frame. The modulator modulates the processed digitized speech information frame onto an RF carrier for subsequent transmission through the antenna.

Brief description of the drawings

All features and advantages of this invention will become readily apparent to those of ordinary skill in the art from the following written description when taken in conjunction with the accompanying drawings.

Figure 1 is a general functional block diagram of the invention.

Figure 2 illustrates the frame and position structure of the US Digital Standard IS-54 for cellular radio communications;

Figure 3 is a block diagram of a first preferred embodiment of the invention implemented using a digital signal processor;

Fig. 4 is a functional block diagram of an exemplary embodiment of the invention as applied to one of a plurality of portable radio transceivers in a radio communication system.

5A and 5B are a flowchart illustrating the functions/operations performed by the digital signal processor in implementing the first preferred embodiment of the invention.

Fig. 6A is a first exemplary diagram illustrating the attenuation-frequency characteristic of the filter circuit according to the first preferred embodiment of the present invention.

Fig. 6B is a second exemplary diagram illustrating the attenuation-frequency characteristic of the filter circuit according to the first preferred embodiment of the present invention.

Figure 7 is an exemplary look-up table accessible by the filter controller circuit of the first preferred embodiment of the invention.

8A and 8B illustrate amplitude-frequency characteristics of an example input audio signal.

Figures 9A and 9B illustrate the amplitude-frequency characteristics of the input audio signal in Figures 8A and 8B after being filtered by the inventive filter circuit, respectively;

Figure 10 is a block diagram of a second preferred embodiment of the invention implemented using a digital signal processor;

Fig. 11 is a flowchart, corresponding to the flowchart of Fig. 5B, which illustrates the functions/operations performed by the digital signal processor in implementing the second preferred embodiment of the invention.

Figure 12 is an exemplary look-up table accessible by the filter control circuit in the second preferred embodiment of the invention.

Detailed description of the drawings

In the following description, for purposes of explanation rather than limitation, specific details are set forth such as particular circuits, circuit elements, techniques, flowcharts, etc., in order to give a thorough understanding of the invention. It will be apparent, however, to those skilled in the art that the invention may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known methods, devices, and circuits are omitted so as not to obscure the description of the invention with unnecessary detail.

FIG. 1 is a generalized block diagram of an adaptive noise reduction system 100 according to the invention. The adaptive noise reduction system 100 includes a filter control circuit 105 connected to a filter circuit 115 . Filter control circuit 105 generates a filter control signal for the current frame of the digitized audio signal. The filter control signal is output to filter circuit 115 which is adjusted in accordance with the filter control signal to exhibit a high pass frequency response curve selected on the basis of the filter control signal. The conditioned filter circuit 115 filters the current frame of the digitized audio signal. The filtered signal is processed by a sound encoder 120 to produce an encoded signal representing the digitized audio signal.

In an exemplary implementation of the invention as applied to a portable/mobile radiotelephone transceiver in a cellular radio communication system, Figure 2 illustrates the Time Division Multiple Access (TDMA) frame adopted by the IS-54 standard for digital cellular radio communication structure. A "frame" is a 20 millisecond time period that includes a transmit block TX, a receive block RX and a signal strength measurement block for Mobile Assisted Handover (MAHO). The two consecutive frames shown in Figure 2 are sent over a period of 40 milliseconds. The digitized speech and background noise information is processed and filtered on a frame-by-frame basis as described further below.

Preferably, the functions of the filter control circuit 105, the filter circuit 115 and the voice coder 120 in FIG. 1 are realized by a high speed digital signal processor. A suitable digital signal processor is the TMS320C53 DSP available from TI Corporation. The TMS320C53 DSP includes a 16-bit microprocessor on a single integrated chip, on-chip RAM for storing data such as speech frames to be processed, and ROM for storing various data processing algorithms, including VSELP speech The compression algorithm, as well as other algorithms described below, are used to complete the functional blocks performed by filter control circuit 105 and filter circuit 115 .

A first embodiment of the invention is represented in FIG. 3 . In a first embodiment, the filter circuit 115 is adjusted as a function of the background noise estimate. The background noise estimate is determined by a filter control circuit. Pulse code modulated audio frames are sequentially stored in the on-chip RAM of the DSP. Other digitization techniques may be used to digitize the audio information. Each PCM digital frame is fetched from the DSP on-chip RAM and processed by the frame energy estimator 210, and then temporarily stored in the temporary frame memory 220. The energy of the current frame determined by the frame energy estimator 210 is provided to the noise estimator 230 and speech detector 240 functional blocks. When the frame energy estimate exceeds the sum of the previous noise estimate and a speech threshold, the speech detector 240 indicates that speech is present in the current frame. If the speech detector determines that no speech is present, the digital signal processor 200 calculates a revised noise estimate as a function of the current noise estimate and the energy of the current frame.

The corrected noise estimate is output to the filter selector 235 . Filter selector 235 generates a filter control signal based on the noise estimate. In a preferred embodiment, filter selector 235 reads a look-up table during generation of the filter control signal. The look-up table includes a series of filter control values, each control value matching a noise estimate or range of noise estimates. On the basis of the corrected noise estimate, a filter control value is selected from the look-up table, the filter control value being represented by a filter control signal which is output by the filter circuit 115 to the filter bank 265 . To stabilize the process and avoid continuous switching between different filters, a switching time of N frames is set for the selection of a new filter. A new filter can only be selected once every N frames, where N is an integer greater than 1, and preferably greater than 10.

The filter circuit 115 is adjusted according to the filter control signal to exhibit a high-pass frequency response curve corresponding to the input filter control signal and the noise estimate. Various types of filter circuits well known in the art can be used to display a selected frequency response curve based on the filter control signal. These prior art filters include IIR filters such as Butterworth, Chebyshev or elliptic filters, FIR filters can also be used due to lower processing requirements, but IIR filters are preferred.

The filtered signal is processed by a vocoder 120 which is used to compress the bit rate of the filtered signal. In a preferred embodiment, the audio encoder 120 encodes the audio signal using Vector Source Codebook Excited Linear Predictive Coding (VSELP) techniques. Other audio coding techniques and algorithms can also be used, such as Code Excited Linear Prediction (CELP) coding, Residual Pulse Excited Linear Prediction (RPE-LTP) coding, Improved Multiple Band Excitation (IMBE) coding. Background noise is minimized by filtering the audio signal frames according to the invention prior to vocoding, a process that substantially cuts out any unwanted noise effects in the speech when it is reconstructed. It also prevents speech from being "swamped" in low frequency noise.

The digital signal processor 200 described in connection with FIG. 3 can be used in a device such as a transceiver of a digital portable/mobile radiotelephone used in a radio communication system. Figure 4 illustrates one such digital radio transceiver which may be used in a cellular radio communication network.

An audio signal including speech and background noise is input from a microphone 400 to a codec 402, which is preferably an application specific integrated circuit (ASIC). The band-limited audio signal detected at microphone 400 is sampled by codec 402 at a sampling rate of 8000 samples per second and blocked into frames. According to the above, each 20 millisecond frame includes 160 speech samples. These samples are quantized and converted into a coded digital format such as 14-bit linear PCM. Once the 160 samples of the digitized speech of the current frame are stored in the on-chip RAM in the transmitting DSP 200, the transmitting DSP 200 performs channel coding functions, frame energy estimation, noise estimation, speech detection, FFT, filtering functions and digital speech coding/compression.

Supervisory microprocessor 432 controls the overall operation of all components in the transceiver shown in FIG. The filtered PCM data stream generated by transmit DSP 200 is provided for quadrature modulation and transmission. So far, based on the filtered PCM data stream from DSP 200, ASIC gate array 404 generates an in-phase (I) information channel and a quadrature (Q) information channel. The I and Q bit streams are processed by matched low pass filters 406 and 408 and passed to the IQ mixer in balanced modulator 410 . Reference oscillator 412 and multiplier 414 provide a transmit intermediate frequency (IF). The I signal is mixed with the in-phase IF, and the Q signal is mixed with the quadrature IF (ie, the in-phase IF is delayed by 90 degrees by phase shifter 416). The mixed I and Q signals are summed and "up"converted to a selected RF channel frequency by channel combiner 430 and then transmitted via duplexer 420 and antenna 422 on the selected radio frequency channel.

On the receive side, the signal received via antenna 422 and duplexer 420 is down-converted from the selected receive channel frequency in mixer 424 to a first IF frequency using The output of reference oscillator 428 is based on the synthesized local oscillator signal. The output of the first IF mixer 424 is filtered and its frequency down-converted to the second IF, which conversion is performed on the basis of the other output of the channel combiner 430 and demodulator 426 . Receive gate array 434 then converts the second IF signal into a series of phase samples and a series of frequency samples. Receive DSP 436 performs demodulation, filtering, gain/attenuation, channel decoding and speech expansion on the received signal. The processed speech data is then sent to the codec 402 and converted to a baseband audio signal for driving the speaker 438 .

The operations performed by the digital signal processor 200 to realize the functions of the filter control circuit 105, the filter circuit 115 and the voice encoder 120 will now be described with reference to the flowcharts in FIGS. 5A, 5B. The frame energy estimator 210 determines the energy of each frame of the audio signal. The frame energy estimator 210 determines the energy of the current frame by computing the sum of the squared values of each PCM sample in a frame (step 505). Since for a sampling rate of 8000 samples per second, there are 160 samples per frame that is 20 milliseconds long, then 160 PCM sample squares are summed. Expressed mathematically, the frame energy estimate is determined according to Equation 1 below: Equation 1

The frame energy value calculated for the current frame is stored in on-chip RAM 202 of DSP 200 (step 510).

The speech detector 240 function includes fetching a noise estimate previously determined by the noise estimator 230 from the on-chip RAM 202 of the DSP 200 (step 515). Of course, no noise estimate exists when the transceiver is initially powered up. Decision block 520 anticipates this and provides a noise estimate at step 525 . In order to force a correction of the noise estimate as will be described below, it is preferable to assign an arbitrarily high value for the noise estimate, for example 20dB above the normal speech level. The frame energy determined by frame energy estimator 210 is fetched from on-chip RAM 202 of DSP 210 (block 530). In block 535 it is determined whether the frame energy estimate exceeds the sum of the detected noise estimate plus a predetermined speech threshold, as expressed in Equation 2 below:

Frame energy estimate > (noise estimate + speech threshold) (Equation 2)

The speech threshold may be a fixed value empirically determined to be greater than the short-term energy variance of general background noise, and may be set to, for example, 9dB. Additionally, speech thresholds can be adaptively modified to reflect changing speech conditions, for example, when a speaker enters a noisier or quieter environment. If the frame energy estimate exceeds the sum in Equation 2, then a flag is set in block 570 to indicate the presence of speech. If the speech detector 240 detects the presence of speech, the noise estimator 230 is bypassed and the noise estimate calculated for the previous digitized audio frame is retrieved and used as the current noise estimate. Conversely, if the frame energy estimate is less than the sum in Equation 2, the speech flag is cleared at block 540 .

Other systems that detect speech in the current frame can also be used. For example, the European Telecommunications Standards Institute (ETSI) has developed a standard for Voice Activity Detection (VAD) in the Global Positioning System GSM. and is described in ETSI reference: RE/SMG-020632P, which is hereby incorporated by reference.

If speech is not present, the noise estimate modification routine in noise estimator 230 is executed. During times when no speech is present, the noise estimate is essentially an online average of the frame energy. As described above, if the initial start-up noise estimate is chosen high enough, speech is not detected, and the speech flag is therefore cleared to force a revision of the noise estimate.

In the noise estimation routine performed by the noise estimator 230, a difference/error (Δ) is determined in block 545, which is the difference between the frame noise energy produced by the frame energy estimator 210 and the noise The difference between noise estimates previously computed by estimator 230:

Δ = current frame energy - previous noise estimate (Equation 3)

Decision block 550 determines if Δ exceeds zero. If Δ is negative, as occurs with high noise estimates, then the noise estimate is recalculated in block 560 according to the following equation:

Noise Estimate = Previous Noise Estimate + Δ/2 (Equation 4)

Since Δ is negative, this causes the noise estimate to be corrected downward. Here a relatively large step size Δ/2 is chosen for fast correction to reduce the noise level. However, if the frame energy exceeds the noise estimate, giving a Δ greater than 0, then in block 555 the noise is corrected according to the following equation:

Noise Estimate = Previous Noise Estimate + Δ/256 (Equation 5)

Since Δ is positive, the noise estimate must increase. However, here a smaller step size Δ/256 (compared to Δ/2) is chosen to gradually increase the noise estimate and substantially eliminate the instantaneous noise.

The noise estimate calculated for the current frame is output to the filter selector 235 . In the first preferred embodiment, filter selector 235 reads the look-up table and uses the current noise estimate to select a filter control value (step 572). Filter circuit 115 (step 574) is then adjusted as a function of the selected filter control values to exhibit a frequency response curve intended to increase the amount of noise filtered as the noise estimate and background noise increase. The PCM samples stored in DSP RAM are then filtered by adjusted filter circuit 265 to remove noise (step 576). The filtered PCM samples are then processed by the vocoder 120 (step 578), and the encoded samples are then output to the RF transmit circuit (step 580).

6A and 6B give several examples of how the filter circuit can be adjusted to exhibit different frequency response curves F1 - F4 for different filter control signals input to the filter circuit 115 . As shown in FIG. 6A, the filter circuit 115 can be selected to exhibit a series of different frequency response curves, and the frequency response curves F1-F4 have cut-off frequencies F1c-F4c, respectively. In a preferred embodiment, the cutoff frequency of filter circuit 115 may range from 300 Hz to 800 Hz. As the noise estimate increases, the filter circuit 115 is designed to exhibit a frequency response curve with a higher cutoff frequency. This higher cutoff frequency results in a greater portion of the frame energy falling in the low frequency range of speech being extracted by filter circuit 115 .

Likewise, as shown in FIG. 6B, the filter circuit 115 can be selected to display a series of different frequency response curves F1-F4, and each frequency response curve has a different slope and the same cutoff frequency. The cutoff frequencies of the frequency response curves F1-F4 are within the range mentioned above. As the noise estimate increases, the filter circuit 115 is adjusted to exhibit a frequency response curve with a steeper slope. This steeper slope results in a greater fraction of the frame energy falling in the lower frequency range of speech being extracted by filter circuit 115 .

Filter circuit 115 filters the current frame as a function of some noise estimate calculated for the current frame. The current frame is filtered so that the noise is cut and the main part of the speech is passed. The main part of the speech that is not filtered and passed gives a recognizable speech output with only a small degradation in speech signal quality. Combinations of different cutoff frequencies and different slopes can be used to adaptively extract selected portions of frame energy that fall within the low frequency range of speech.

FIG. 7 depicts an example look-up table read by filter selector 235 to select one of filter response curves F1-F4 for filter circuit 115. Referring to FIG. The look-up table includes a series of possible noise estimate values N1-Nn and filter control values F1-Fn corresponding to possible response curves displayed by the filter circuit 115 . Each of the noise estimates N1-Nn may represent a range of noise estimates, and each is matched to a particular filter control value F1-F4. Filter control circuit 105 generates a filter control signal by computing a noise estimate and retrieving the associated filter control value from a look-up table.

Figures 8A&B and 9A&B illustrate how each of the two frames of the audio signal is adaptively filtered to give an improved audio signal output to the RF transmitter. 8A and 8B show a first frame and a second frame of an audio signal including speech components s1, s2 and noise components n1, n2, respectively. As shown, the noise energies n1 and n2 in both frames are concentrated in the low audio frequency range. The speech energies s1 and s2 are concentrated in the higher audio frequency range. FIG. 9A shows the noise signal n1 and the speech signal s1 of the first frame after filtering. FIG. 9B shows the noise signal n2 and the speech signal s2 of the second frame after filtering.

As discussed, the adaptive audio noise reduction system 100 is designed to calculate the difference in noise level between a first frame and a second frame by adjusting the filter control based on the calculated noise estimate for the current frame circuit 105. For example, filter control circuit 105 computes noise estimate N1 and spectral envelope s1 and selects a filter control value F1 for the first frame. In the preferred embodiment, based on the filter control value F1, the filter circuit 115 is tuned and exhibits a frequency response curve having a cutoff frequency F1c as shown in FIG. 6A. Then, the first frame passes through the adjusted filter circuit 115 . The filter circuit 115 is selected such that most of the noise n1 and only a small part of the speech s1 fall below the cut-off frequency F1c of the frequency response curve F1. This results in the noise n1 being effectively filtered out and only a part of the relatively unimportant speech s1 being filtered out. The filtered audio signal of the first frame is shown in Fig. 9A.

In the second frame shown in FIG. 8B , there is higher background noise, and assuming speech is not detected, the filter control circuit 105 calculates a higher noise estimate n2. Based on the higher noise estimate, a higher corresponding filter control value F2 is determined for the second frame. In the first preferred embodiment, the filter circuit 115 is adjusted according to the higher filter control value F2 to exhibit a frequency response curve with a higher cutoff frequency F2c as shown in FIG. 6A. Subsequent frames of the audio signal then pass through the adjusted filter circuit 115 . Since the cut-off frequency F2c of the frequency response curve F2 is higher for subsequent frames, most of the noise n2 and speech s2 are filtered out. (However), the filtered out part of the speech s2 is still relatively insignificant compared to the intelligibility information included in the frame, so this has only a small impact on the speech. The disadvantage of filtering out a larger portion of the speech s2 is offset by the advantage of increased noise n2 removal in the second frame. The portion of the speech spectrum that is filtered out does not significantly contribute to speech intelligibility. The filtered audio signal in the second frame is shown in Fig. 9B.

A second preferred embodiment of an adaptive noise reduction system 100 is shown in FIGS. 10-12. In a second preferred embodiment, filter control circuit 105 adjusts filter circuit 115 as a function of the noise envelope estimate. A noise profile estimate is calculated for each frame and compared to a reference noise profile estimate. Based on this comparison, filter circuit 115 is adaptively adjusted to extract different amounts of low frequency energy from the current frame.

Referring to Figure 10, a DSP 200 configured according to a second preferred embodiment is shown. As shown, filter control circuit 105 includes spectral analyzer 270 in addition to frame energy estimator 210, noise estimator 230, speech detector 240 and filter selector 235 described with reference to the first preferred embodiment. As described for the first embodiment and represented by flowcharts 5A and 5B, filter control circuit 105 determines noise estimates for received frames and detects the presence of speech. The spectral analyzer 270 modifies the noise envelope estimate and uses this value in the adjustment filter circuit 115 when detecting speech for the current frame.

Referring to FIG. 11 , the steps of modifying the noise profile estimate and adjusting the filter circuit 115 are given. Figure 11 shows the steps performed by spectrum analyzer 270, which are referenced throughout the process previously described in flow diagrams 5A and 5B of the first preferred embodiment.

If no speech is detected in the current frame, spectrum analyzer 270 first determines a noise profile for the current frame (step 600). The noise profile determined for the current frame includes calculated energy values at different frequencies (ie, frequency bins) that are located in the speech low frequency range selected for the current frame. In a preferred embodiment, the selected frequency range is approximately 300 to 800 Hz. The noise profile of the current frame can be determined by processing the current frame with a Fast Fourier Transform (FFT) with N frequency bins. Using FFT to process digital signals is well known in the art. It has the advantage that when FFT is limited to relatively few frequency points, eg 32 points, it requires little processing power. An FFT with N frequency bins produces energy calculations at N different frequencies. The calculated energies of frequency points falling within the selected frequency range form the noise envelope for the current frame.

To determine a noise profile estimate for the current frame (step 604), the noise profile for the current frame is averaged with noise profile estimates determined for previous frames of the audio signal. The stored initial noise profile estimate may be used when no previous noise profile estimate is available, eg after initialization. The _noise envelope estimate consists of noise energy estimates _{e i (} where i=1, 2...n). In a preferred embodiment, each noise energy estimate _ei corresponds to the average value of the energy calculations at a specific frequency, which is a selected frequency range over a large number of consecutive frames in which no speech is detected A frequency point within . By using a large number of frames to determine the noise profile estimate, the filter circuit 115 is adjusted on a more gradual basis. In another embodiment, the noise profile estimate may be equal to the noise profile of the current frame.

Then, the energy estimate _ei of the noise profile estimate is compared with the reference noise profile (step 604). The reference noise profile includes reference energy thresholds e _ri (where i=1, 2...n) at frequency bins corresponding to noise energy estimates _ei of the noise profile estimates. The reference energy threshold e _ri can be empirically determined. In order from the highest frequency energy estimate e ₁ to the lowest frequency energy estimate e _n , the noise energy estimates e _i are successively compared with corresponding reference energy thresholds e _ri .

More specifically, the noise energy estimate e ₁ is first compared with a reference noise threshold e _r1 . If e ₁ is greater than the reference noise threshold e _r1 , the comparison value C1 is selected and input to the filter selector 235 . If the noise estimate e ₁ is smaller than the reference noise threshold e _rl , then the noise energy estimate e ₂ , which is the noise energy estimate obtained at a frequency lower than e ₁ , is compared with the reference threshold e _r2 . If the noise energy estimate e ₂ is greater than the reference noise threshold e _r2 , the comparison value C ₂ is selected and input to the filter selector 235 . The comparison process continues until a comparison value C _i (where i=1, 2...n) is selected.

The filter circuit 235 uses the determined comparison value Ci to determine a filter control value. The filter control values are selected from a look-up table such as that given in FIG. 12 . The look-up table includes a series of comparison values Ci and corresponding filter control values Fi. The filter circuit 115 is adjusted as a function of the selected filter control value. Filter circuit 115 is tuned to display a frequency response curve to extract low frequency energy from the current frame. When the noise energy estimates at successively higher frequencies exceed their corresponding reference energy thresholds, the filter circuit 115 is adjusted to extract more low frequency energy. Figures 6A and 6B show example frequency response curves for selected filter control values.

The use of the noise envelope estimate helps improve the ability to adaptively adjust the filter circuit to extract low frequency energy in a way that helps improve the overall quality of speech. Since the automotive environment is not the only environment in which mobile radios are used. Therefore, the noise profile in an environment may be skewed toward higher frequencies. Spectrum analyzer 270 can be selectively disabled when the noise energy in low frequencies is small. Also, when a significant portion of the noise frequency spectrum is at low frequencies, then steeper filtering slopes are applied even though some processing power is sacrificed. This additional processing requirement is still minimal.

As apparent from the above description, the adaptive noise filtering system of the invention is simply implemented. And the calculation amount of DSP does not increase significantly. More complex methods of noise reduction, such as "spectral reduction" require several MIPS involved in the computation and a large amount of memory for storing data and program code. By comparison, the invention can be implemented using only a fraction of the MIPS and memory required by the "spectral clipping" algorithm, which also introduces more speech distortion. Reduced memory reduces the size of the DSP integrated circuit; reduced MIPS reduces power consumption. These features are ideal for battery-operated portable/mobile radiotelephones.

While the invention has been particularly shown and described with reference to its preferred embodiments, it is not limited thereto. For example, although the DSP is described as performing the functions of frame energy estimator 210, noise estimator 230, speech detector 240, filter selector 235 and filter circuit 265, these functions may be implemented using other digital and/or analog components to fulfill. Furthermore, the adaptive filtering system 100 can also be implemented when the filter circuit 115 is adjusted as a function of both the noise estimate and the noise envelope estimate.

Claims (10)

1. method that is used for optionally changing a frame of digital signal, digital signal wherein is made of a plurality of successive frames, the sound signal that this digital signal representation receives at the transmitter place, this sound signal or constitute by speech components, or constitute by noise component, perhaps be made of jointly speech components and noise component, described method is characterised in that and may further comprise the steps:

The energy sizes values of estimative figure signal frame;

Determine whether comprise speech components in the digital signal frame according to the estimated value that obtains in the described estimating step;

When determining that in described determining step speech components does not constitute frame a part of, revise noise estimation value with the form of the energy sizes values function estimated in previous noise estimation value and the described estimating step;

Read a record in the question blank, this question blank has the filter characteristic according to the noise estimation value size index, and the record of reading is corresponding to the noise estimation value of revising in the described correction step;

Select to want the filter circuit filtering characteristic of filtered device demonstration, the filtering characteristic of choosing is corresponding to the filter characteristic of storage in reading to note down in the described read step;

Use the filter filtering digital data frames, wave filter wherein shows the filter circuit filtering characteristic, changes digital data frames according to the filter circuit filtering characteristic thus.

2. according to the method in the claim 1, its feature is also the intermediate steps of adding not comprise speech components if digital data frames is determined that this additional step is determined the noise envelope estimated value of digital signal frame so.

3. according to the method in the claim 2, the noise envelope estimated value that wherein is determined in described definite noise envelope estimated value step is used to revise noise estimation value in described correction step.

4. according to the method in the claim 1, wherein question blank is read in described read step, and it is characterized in that has a lot of records in the look-up table, and each record all comprises an independently filter characteristic.

5. according to the method in the claim 4, wherein the separate filter characteristic of a plurality of records comprises independently high pass filter characteristic in the question blank, and each high-pass filtering characteristic is by an independent cutoff frequency definition.

6. according to the method in the claim 4, wherein the separate filter characteristic of a plurality of records comprises independently high pass filter characteristic in the question blank, and each high-pass filtering characteristic is defined by an independent frequency response curve gradient.

7. according to the method in the claim 1, it is characterized in that further step: thus Counter Value statistics frame number of estimated energy sizes values for it in described estimating step increased.

8. according to the method in the claim 7, when every N counting value increases, just carry out the step of described selective filter circuit median filter characteristic, N be one greater than 1 integer.

9. device (100 that is used for optionally changing digital signal frame; 200), digital signal wherein is made of a plurality of successive frames, the sound signal that this digital signal representation receives at the transmitter place, this sound signal can be made of speech components, or constitute by noise component, perhaps constituting jointly by these two components, described device is characterised in that:

Be coupled the energy value estimator (210) of receiving digital signals frame identification, described energy value estimator is used for the energy value of estimative figure signal frame;

Be coupled to the speech detector (240) of described energy value estimator, described speech components determiner is used for determining whether digital signal frame comprises speech components;

Determine not exercisable noise estimator (230) during configuration frame a part of of speech components when described speech components determiner, described noise estimator is revised noise estimation value with the form of the energy value function that front noise estimation value and described estimator are estimated;

The question blank that comprises a lot of records, wherein every record comes index with noise estimation value, and the record in the question blank reads according to the noise estimation value that is formed by described noise estimator;

Be coupled the wave filter (265) of receiving digital data frame, described wave filter shows the filtering characteristic of selectable filter circuit, selection to the filter circuit filtering characteristic of wave filter is to determine according to the record of question blank, and the record of this question blank reads according to the noise estimation value by described noise estimator correction.

10. according to the device in the claim 9, its feature also is a noise envelope estimator (270), if described speech components determiner determines that digital data frames does not comprise speech components, this noise envelope estimator is determined the noise envelope estimated value of this digital data frames so.

Images