KR20040075959A - Voice activity detector and validator for noisy environments - Google Patents

Abstract

The communication unit 100 includes an audio processing unit 109 having voice activity detection mechanisms 130, 135. Voice activity detection mechanisms 130 and 135 measure the energy acceleration of the signal input to communication unit 100 to determine whether the input signal is negative or noise based on the measurement. A speech detection method and a method of determining whether an input signal is sound or noise are also described. In particular, the use of energy acceleration based on voice activity detectors and validators for a noisy environment offers the advantages of independence of the input sound level, fast response and noise robustness.

Description

Voice activity detector and validator for noisy environments

Many voice communication systems, such as the TETRA (TErrestrial Trunked RAdio) system and the Global Mobile Communication System (GSM) cellular telephone standard for private mobile wireless users, use sound processing units to encode and decode sound patterns. . In such voice communication systems, the sound encoder converts the analog voice pattern into an appropriate digital format for transmission. The tone decoder converts the received digital tone signal into an audible analog tone pattern.

Methods and apparatus for detecting voice activity are known in the art. Voice activity detector (VAD) operates under the assumption that sound is provided only in the audio signal portion. This assumption is usually correct because there are many audio signal intervals that merely indicate silence or background noise.

Voice activity detectors can be used for many purposes. When no sound is present, these uses include suppressing the overall transmission activity in the transmission system, thereby significantly saving power and channel bandwidth. When the VAD detects that the tone activity has resumed, it reinitializes the transmission activity.

By distinguishing audio portions that contain sound from those that are “speechless”, voice activity detectors can also be used in combination with sound storage devices. Thereafter, the portions containing the sound are stored in the storage device and the "silent" portions are discarded.

Conventional methods of detecting speech are based at least in part on methods of detecting and evaluating the power of a sound signal. The estimated power is compared with one of the constant or adaptive thresholds to determine if the signal is negative. The main advantage of these methods is their low complexity, making them suitable for implementing low-processing resources. The main disadvantage of such methods is that background noise can be inadvertently generated in the "sound" detected when the "sound" is not actually provided. Alternatively, the "sound" provided is unclear because it is unclear and difficult to detect due to background noise.

Some methods for detecting sound activity relate to noise transfer environments and are based on adaptive filtering of the sound signal. This reduces the amount of noise from the signal before the final decision. The frequency spectrum and noise level can be varied because this method is used in several speakers and in various environments. Therefore, input filters and thresholds are often adaptive, tracking these changes.

Examples of these methods are provided in the GSM Specification 06.42 VoiceActivity Detector (VAD) for each of the half rate, full rate and enhanced full rate sound traffic channels. Another such method is the "Multi-Boundary Voice Activity Detection Algorithm" as proposed in ITU G.729 Annex B. These methods are very accurate in noisy environments but are quite complex to perform.

All these methods require a sound signal to be input. Some applications using the sound decompression scheme require performing sound detection during the sound decompression process.

European patent application EP-A-0785419 by Benyassine et al. Relates to a method for voice activity detection comprising the following steps.

(i) extracting a set of predetermined parameters from the incoming voice signal every frame; And,

(ii) making frame speech determination of the incoming sound signal every frame according to a set of difference measures extracted from the set of predetermined parameters.

In cellular systems, the VAD is biased to enable the radio, including sound codecs and RF circuitry, to communicate to the other party in the presence of background noise and other damages when the party speaks. However, it transmits data even when the parties are not talking. This slightly lowers battery life and slightly increases interference with common-channel users in other cells of the system. These are essentially secondary (or higher) effects.

In these systems, there is no concept of finite resources being made available for duplex calls. This is possible and consistent for the uplink and the downlink as a whole, which is typically present on multiple carriers to make use of the full bandwidth simultaneously.

In the field of the present invention, some speech activity or speech onset detectors (VADs / VODs) attempt to distinguish speeched speech using sound characteristics such as harmonic structure (eg, via autocorrelation). Known. However, under noise, these structural indicators may fail due to either destruction of the sound structure or noise structure. This may be, for example, the engine, tire, or air conditioner noise of the vehicle. Finally, these methods are not good for detecting unvoiced sounds.

An alternative method is to just use the frame energy level to detect sound. This is satisfactory for the sound of high signal-to-noise ratio (SNR) states, where any threshold above the noise level can be set to indicate the sound. However, this method fails in more practical noise states.

For non-standardized databases or in practical applications, the noise levels in one set of examples may be greater than the sound levels in other examples, which makes it impossible to set a threshold. A common way to overcome this is to average the first 100 msec of utterance under the assumption that it represents noise, to produce a specific threshold for this pronunciation. However, again, this may result in noise being radiated rapidly from the initial estimate, and non-fixed noise containing noise rather than noise having a high variance or the first few frames are not actually estimated noise. It is insufficient for stationary noise.

Therefore, there is a need for improved voice activity detectors and valley data for noise environments that can alleviate the above-mentioned disadvantages.

The present invention relates to sound detection (typically known as voice activity detection (VAD)) within a noisy environment. The invention applies to, but is not limited to, energy acceleration measurements of speech signals in a sound detection system.

1 is a block diagram of a communication unit adapted to perform voice activity detection and detection of a preferred embodiment of the present invention.

2 is a flow chart of energy acceleration based on voice activity detector for noise environments in accordance with a preferred embodiment of the present invention.

3 is a flow chart of energy acceleration based on voice activity verification for noise environments in accordance with a preferred embodiment of the present invention.

Figure 4 illustrates a buffer operation in accordance with a preferred embodiment of the present invention.

According to a first aspect of the invention, there is provided a communication unit as claimed in claim 1.

According to a second aspect of the invention, a method is provided for detecting a sound signal input to a communication unit as claimed in claim 11.

According to a third aspect of the invention, a method is provided for determining whether a signal input to a communication unit as claimed in claim 14 is negative or noise.

Additional aspects of the invention are as claimed in the dependent claims.

In summary, the present invention seeks to address the case of unfixed noise of arbitrary amplitude by using energy acceleration measurement rather than measuring energy amplitude to indicate the presence or absence of sound.

Exemplary embodiments of the invention will now be described with reference to the accompanying drawings.

Since the onset of the spoken sound depends on the activity of the vibrating or stationary vocal cords, the spoken sound has a relatively high-energy acceleration value. Similarly, non-negative onsets (eg, burst sounds) also have high energy accelerations.

The inventors have recognized that in a representative domain stressing voicing, such as narrowband power spectrum or Mel-spectrum, synthetic energy acceleration is significantly higher than unfixed noise. Only important exceptions are irritating noises (eg, applause).

Therefore, in accordance with preferred embodiments of the present invention, the inventors have recognized that they can additionally discriminate against these noises by concentrating on energy in the frequency domain likely to include the basic pitch of the speech signal. In particular, the inventors of the present invention have proposed using negative unstructured properties, i.e., energy properties (or acceleration of any metric that reflects negative energy or components thereof).

In particular, a preferred application to the concepts of the invention described herein is the European Telecommunications Standards Institute (ETSI)-"Speech Processing, Transmission and Quality aspects (STQ): Distributed speech recognition; Front-end feature extraction algorithm; Compression algorithm" , Distributed speech, currently defined in ETSI ES 201 108 v1.1.2 (2000-04), April 2000.

Referring now to FIG. 1, a block diagram of an audio subscriber unit 100 is shown to support the inventive concept of preferred embodiments of the present invention.

A preferred embodiment of the present invention is, for example, 3GPP (3) for future cellular wireless communication systems.^rd Generation Partnership Project) is described in connection with a wireless audio communication unit that can operate as a standard to provide DSR capabilities. However, the concepts of the present invention described herein in connection with voice activity detection and verification can be equally applied to any electronic device responsive to voice signals and benefit from improved voice activity detection circuitry. It is considered in the present invention.

As is known in the art, the audio subscriber unit 100 is a circuit that isolates between receive and transmission chains within an antenna 102, antenna switch or audio subscriber unit 100, preferably coupled to a duplex filter. A circulator 104.

The receiver chain includes a receiver front-end circuit 106 (which efficiently provides reception, filtering and intermediate or baseband frequency conversion). The front-end circuit 106 is coupled in series to the signal processing function (typically implemented with a digital signal processor (DSP)) 108. The signal processing function 108 performs signal demodulation, error correction, and formatting. The data recovered from the signal processing function 108 is serially coupled to the audio processing function 109, which formats the received signal in an appropriate manner to format the audio pronunciation / display. To 111.

In various embodiments of the invention, the signal processing function 108 and the audio processing function 109 may be provided in the same physical device. The controller 114 is configured to control the operational state and information flow of the elements of the subscriber unit 100.

With respect to the transmission chain, this is essentially an audio input device 120 coupled in series via the audio processing function 109, the signal processing function 108, the transmitter / modulation circuit 122 and the power amplifier 124. ). Processor 108, transmitter / modulation circuitry 122, and power amplifier 124 are operatively responsive to the controller. The power amplifier output is coupled to a duplex filter, an antenna switch or circulator 104 and an antenna 102 that radiates the final radio frequency signal.

In particular, the audio processing function 109 includes a voice activity (or voice onset) detection (VAD) function 130 operatively coupled to the voice activity determining function 135. According to preferred embodiments of the present invention, the VAD function 130 and the voice activity determination function 135 are adapted to provide an improved voice detection and determination mechanism, the operation of which is related to FIGS. 2 and 3. The addition is explained. Voice activity detector function 130 includes a frame-by-frame detection stage every frame of the three measurements. Three frequency range measurements include:

(i) full spectrum;

(ii) spectral sub-bands; And,

(iii) spectral variance

Next, the voice activity determination function 135 makes a decision based on a buffer of measurements, which are analyzed for their speech likelihood. The final decision from the decision stage is applied retrospectively to earlier frames in the buffer.

In a preferred embodiment of the present invention, timer / counter 118 is also adapted to perform timing functions in the detection and determination processes of FIGS. 2 and 3.

The signal processor function 108, the audio processing function 109, the VAD function 130, and the voice activity determination function 135 may be implemented as separate operatively coupled processing elements. Alternatively, one or more processors may be used to perform one or more of the corresponding processing operations. In yet another additional embodiment, the above-described functional units may be implemented as a combination of hardware, software or firmware elements, using application specific semiconductors (ASICs) and / or processors, for example digital signal processors (DSPs). Can be.

Of course, the various components in the audio subscriber unit 100 can be realized in discrete or integrated component form, so that the final structure is made only by arbitrary choice.

As such, there are several ways to achieve an indication of energy acceleration for use in the preferred embodiment of the present invention.

(i) The theoretical ideal method is to double-differentiate the energy level over successive frames of pronunciation, as can be seen in the previously published application US 6009391. The disadvantage of this approach is that it introduces delay because it needs to analyze the number of frames on every frame side under analysis.

(ii) The zero-delay estimate of energy acceleration can be obtained, for example, by comparing the ratio of the instantaneous value to the short term average using the following frame or rolling average.

Frame Average:

Rolling Average:

In each case, this method returns values that can be interpreted as 'deceleration'<'1'<'acceleration'. Then a denominator length that optimally distinguishes the sound from the noise and Experimental values for can be found.

The inventors of the present invention have recognized that the preferred optimal solution is to find a denominator that can track unfixed noise quickly but is too long to track speech onset. The proposed sequence of values for the rolling average is a = 0.2, b = 0.8 * a, c = 0.8 * b, etc., which can be simply expressed repeatedly.

At this time,

Preferred VAD and parameter initialization systems in the detection stage are summarized in the flowchart of FIG. In unfixed noise, long term energy thresholds are negative unreliable indicators. Similarly, in high noise conditions, the structure of a note (eg, harmonics) may not be completely reliable as an indicator because it may be destroyed by noise or structural noises may confuse the detector. Therefore, the preferred sound activity detector is to use the noise-ruggedness characteristic of sound, i.e. energy acceleration associated with speech onset.

Referring now to FIG. 2, a flowchart 200 of a preferred detection process is shown. As mentioned above, this process involves analysis every frame. Preferred VAD mechanisms relate to the 'full spectrum' measurement process. The frame counter is initially evaluated to determine if it is less than 'N' which defines the number of buffered frames as shown in step 205. As an example of the preferred embodiment, if every frame is set to be incremented by 10 msec, 'N' is set to '15'. If the frame counter is less than 'N' in step 205, the rolling average for the initial acceleration test is updated as in step 210. If the frame counter is not less than 'N' in step 205, step 210 is skipped.

Then, as shown in step 235, a determination is made to evaluate whether the energy acceleration measurement is within one or more defined margin (s). If the energy acceleration measurement is within one or more defined margin (s) in step 235, the rolling average is updated with the results of the additional energy acceleration test as in step 240. If the energy acceleration measurement is not within one or more specific margin (s) in step 235, step 240 is skipped.

Thereafter, a determination is made to evaluate whether the energy acceleration measurement is greater than the defined threshold, as shown in step 260. If the energy acceleration measurement is greater than the threshold defined in step 260, the frame is assumed to be a negative frame as in step 265. If the energy acceleration measurement is not greater than the threshold defined in step 260, the frame is assumed to be a noise frame, as in step 270.

The frame counter is then incremented as in step 275 and the process repeats from step 205.

Instead of or in addition to the full spectrum measurement process, as an improvement on the process, the sub-area measurement process shown in optional steps 215 and 245 can be performed. The particular sub-region of the spectrum is selected as most likely to contain the basic pitch.

In the sub-domain process, if the rolling average for the initial acceleration test is updated in step 210 in the full spectral measurement, a determination is made to check whether the energy acceleration measurement is greater than the threshold as shown in step 220. Is done. If the energy acceleration measurement is greater than the threshold in step 220, the initialization process of the other parameters is stopped as shown in step 225. If the energy acceleration measurement is not greater than the threshold in step 220, the initialization of the other parameters is updated as in step 230. The process then returns to step 235 as shown.

An additional preferred decision is made after the decision to evaluate whether the energy acceleration measurement is within one or more defined margin (s) in step 235. The deceleration value is evaluated in step 250 to determine if it is 'high', and if so, the rolling average for the energy deceleration test is slowly updated as shown in step 255. This process then returns to full spectrum method at step 260.

In this way, the generally high signal-to-noise ratios (SNRs) of the sub-region detector make the noise- robustness high. However, it is vulnerable to microphone and speaker changes as well as band-limited noise. Thus, these measurements cannot be trusted in all environments. Consequently, the preferred embodiment of the present invention includes a sub-domain detector to increase the overall spectral measurement.

Additional measurement processes are preferably performed using, for example, the 'acceleration' of the variance of the values in the lower half of the spectrum of every frame. This variance measurement detects the structure in the lower half of the spectrum, making it very sensitive to voiced sound. The variance measurement follows the sub-domain processing scheme, where the lower half of this spectrum is the particular sub-region chosen. This variance measurement further complements the full spectrum measurement approach, allowing better detection of unvoiced sounds and bursting sounds.

All three measurements were originally input from the spectral representation of the filter gains generated by the first stage of a double Wiener filter as described in applicant Motorola, inventor Yan-Ming Chen, US patent application US 09/427497. takes (raw input) As mentioned above, each measurement uses a different aspect of this data.

In particular, the full spectrum detector uses a known Mel-filtered spectral representation of the filter gains generated by the first stage of the double winner filter. The single input value is obtained by square the sum of the Mel filter banks.

In a preferred embodiment of the present invention, the full spectrum detector applies the following process to all frames as described below.

Step 1 initializes the estimated noise tracker (Tracker) in the following manner.

If Frame <15 and Acceleration <2.5, then Tracker = MAC (Tracker, Input).

The energy acceleration measurement prevents the tracker from updating if sound is generated within the lead-in time of 15 frames.

Step 2 updates the tracker value in the following manner if the current input is similar to a noise estimate.

If Input <Tracker * UpperBound and Input> Tracker * LowerBound, then tracker = a * Tracker + (1-a) * Input.

Step 3 provides a safe mechanism for examples where there is loud noise without sound or features within the first few frames. This causes the final errored high catch estimate to decay. Step 3 preferably functions in the following manner.

If Input <Tracker * Floor, Tracker = b * Tracker + (1-b) * Input

Step 4 returns as a 'true' negative decision in the following manner if the current input is greater than 165% above the tracker.

TRUE if Input> Tracker * Threshold, or FALSE otherwise.

The ratio of instantaneous input to short term average tracker is a function of energy acceleration of the continuous inputs.

In the above,

a = 0.8 and b = 0.97

UpperBound is 150% and LowerBound is 75%

Floor is 50%

Threshold is 165%.

Note that if the value is greater than the upper limit or between the lower limit and the floor, it is not updated. In addition, the energy acceleration input can be calculated as one of the following as described above.

Double-differentiation of consecutive inputs, or

Estimated by estimating the ratio of two rolling averages of inputs.

Note that the ratio of the fast and slow-adaptive rolling averages reflects the energy acceleration of successive inputs.

For example, contribution rates for the above-described averages are as follows:

(i) 0 * mean + 1 * input and,

(ii) ((Frame-1) * mean + 1 * input) / Frame,

Energy acceleration measurements increase sensitivity over the first 15 frames.

The sub-band detector preferably uses the average of the derived second, third and fourth Mel-filter banks for the 'full spectrum' measurement. The detector then applies the next process to all frames in the manner described below.

(i) Input = p * CurrentInput + (1- p ) * PreviousInput;

(ii) if Frame < 15, Tracker = MAX (Tracker, Input);

(iii) if Input <Tracker * UpperBound and Input> Tracker * LowerBound, then Tracker = a * Tracker + (1-a) * Input;

(iv) if Input <Tracker * Floor, Tracker = b * Tracker + (1-b) * Input

(v) Output TRUE if Input> Tracker * Threshold, or FALSE otherwise.

In subarea measurements:

p = 0.75.

All other parameters will be the same for the full spectrum measurement except for the threshold, which is the same as 3.25.

For the spectral variance measurement, the variance of the values including the lower frequency half of the narrowband spectral representation of the gain for each frame is used as input. The detector then applies exactly the same process as for the full spectrum measurement.

The variance is calculated as follows.

Where N = FFT Length / 4,

w _i are the values of the narrowband spectral representation of the gain.

According to a preferred embodiment of the present invention, the three measurements described above are provided to the VAD decision algorithm as shown in the flowchart of FIG. Successive inputs are provided to a buffer that provides contextual analysis. This causes a frame delay equal to the length of the buffer minus 1 frame.

Referring now to FIG. 3, a flowchart 300 of an accelerated-based speech activity verification process for noise environments is illustrated in accordance with a preferred embodiment of the present invention.

For an N = 7 frame buffer, the most recent true / false input is stored at position N in the data buffer as shown in step 305. The decision logic applies a number and preferably each of the following steps.

Step 1:

V _N = Measure 1 or Measure 2 or Measure 3

If one of the three measurements returns a true indication, the input (V _N ) is defined as 'true' (T).

Step 2:

The algorithm searches for the longest consecutive sequence of 'true' values in the buffer, as in step 310. Therefore, for example, in the case of the sequence 'T T F T T T F', M becomes equal to '3'.

Step 3:

If M ≧ S _p and T <L _s, then T = Ls.

Here, S _p is equal to the first threshold in step 315. The longest sequence of true (T) negative values is equal to or exceeds the first threshold in step 315, i.e. if S _p = 3 or more consecutive 'true' values, the buffer contains a 'possible' negative. Is determined. If not already provided (or exceeded) from the determination at step 320, a short timer T of L _S = 5 frames Time_1 is activated at step 325.

Step 4:

If M≥S _L and F> F _S, then T = L _M otherwise T = L _L.

Here, S _L is equal to the second threshold in step 330. If S _L = 4 or more consecutive 'true' values are present, it is determined that the buffer again contains 'possible' sounds. As determined in step 355, if the current frame F is outside the initial lead-in safety period F _S , then an intermediate timer T of L _M = 22 frames is activated in step 340. If not, a safe long timer of L _L = 40 frames is used in step 345. This arrangement is used when the earliest existing notes in pronunciation make the initial noise estimate of the VAD too high.

Step 5:

If M <S _P and T> 0, then T--.

If the process determines that at step 350 is less than S _p = 3 consecutive 'true' values and at step 355 the timer is greater than zero, the timer is decremented at step 360.

Step 6:

If T> 0, print TRUE; otherwise, print FALSE. If the timer is greater than zero in step 365, the process outputs a 'true' tone decision as shown in step 370. Alternatively, if the timer is not greater than zero in step 365, the process outputs a 'noise' decision as shown in step 375.

Step 7:

Frame ++, shift the buffer to the left and return to step 1. In preparation for the next frame at step 380, as shown in relation to Figure 4, the buffer is shifted left to accept the next input. Output tone determination is applied to frames ejected from the buffer. This process is then repeated in step 305 for the next true / false input into the data buffer.

It is contemplated to implement alternative mechanisms for making negative or noise determinations based on the energy acceleration process described above. For example, the determination mechanism is not based on one or more timer (s) and may only make a determination if one or more energy acceleration thresholds are exceeded.

Referring now to FIG. 4, an example of a buffer operation 400 in accordance with a preferred embodiment of the present invention is shown in more detail. Assume that the first threshold is set for three consecutive 'true' values. Assume at time 't' 410 that only current input (frame # 7) 425 and pre-input (frame # 6) 420 are 'true'. As a result, when the buffer is shifted, the first frame (frame # 1) 415 will be marked as false.

At time 't + 1' 430, a third 'true' input (frame # 8) 450 is received to supplement the two earlier 'true' inputs 440, 445. Eventually, when the buffer is shifted, the next output frame (frame # 2) will be marked as 'true'.

In the determination process, it should be noted that only the limitations are (i) Time_1 <Time_2 <Time_3 and (ii) Threshold_1 <Threshold_2.

If only these three inputs (frame # 6, frame # 7 and frame # 8) are 'true', the entire output sequence will be as follows.

Here, frames # 2- # 5 represent 'true' due to the buffer read-in function. Frames # 6- # 8 represent 'true' as the location of the actual original 'true' note inputs. Frames # 9- # 12 indicate 'true' due to the buffer read-out function. Frames # 13- # 18 indicate 'true' in response to the timer hangover being used. Once all the frames in the pronunciation are entered, the buffer shifts the 'false' entries (frames # 19- # L _M ) until empty.

It is contemplated in the present invention that the buffer length and hangover timers are dynamically adjusted to meet the needs of the audio communication unit. A preferred embodiment using a buffer length of 'N' of 8 and a hangover timer of 5 frames is used for illustration only. However, the buffer length 'N' must always be determined such that N≥S _L.

In addition to using it as a VAD in the present invention, it is contemplated in the present invention that the energy acceleration measurement performed in the method steps of FIG. 2 can be used to verify the initialization of other parameters. For example, the spectral subtraction scheme requires an initial estimate of noise based on the first negative ten frames (typically 100 msec). Even at fixed noise, several events can occur to invalidate the initial estimate. Examples of such events include the following.

(a) Ramp-up of the signal

For a variety of reasons, the just-initiated record is 'lamped up' to the full volume within the period under evaluation. This is because such an entire ramp-up involves buffer-fill in digital systems, capacitance or tape-head engagement in analog systems. The effect of such events invalidates the estimate. Therefore, energy acceleration measurements can be used to detect such ramp-up and prevent errors.

(b) spikes in the initial signal:

A common 'spike' occurs due to the complete placement of a Press-To-Talk (PTT) button on the subscriber radio unit, where electrical contact occurs before the button strikes the back of the switch. Energy acceleration measurements as described above may be used to stop the estimation process as shown in step 225 of FIG. 2 when such events occur.

(c) the sound at the initial signal:

Another common occurrence with PTT systems is that the user starts talking as soon as the user presses the PTT button. In this way, electrical contact is made after talking. The energy acceleration measurement may identify it as shown in step 255 of FIG. 2 to stop the initializations as noise-based or to use default estimates.

In summary, a communication unit has been described that includes an audio processing unit having a voice activity mechanism. The speech activity detection mechanism provides an indication of energy acceleration of the signal input to the communication unit and determines whether the input signal is sound or noise based on the indication.

In addition, a method of detecting a sound signal input to a communication unit has been described. The method includes indicating acceleration of an input signal to the communication unit; And determining whether the input signal is sound or noise based on the display step.

In addition, a method has been described for determining whether a signal input to a communication unit is negative or noisy. The method includes determining whether the input signal is negative or noise based on energy acceleration, using, for example, a frame average or rolling average of a plurality of input signals.

Therefore, it will be appreciated that the sound activity detector and valley data for the noise environments described above provide the advantages of noise robustness and fast response. Since the preferred embodiment uses measurements according to energy acceleration instead of absolute measurements, the inventive concept described herein can be applied to any input level of sound.

Although specific and preferred implementation manners of embodiments of the present invention have been described above, it will be apparent to those skilled in the art that modifications and variations of the inventive concept are within the scope of the present invention.

Thus, improved sound activity detector and valley data for noise environments can substantially mitigate the aforementioned disadvantages associated with the prior art.

Claims (15)

A communication unit 100 comprising an audio processing unit 109 having a voice activity detection mechanism 130, 135, The voice activity detection mechanisms 130 and 135 measure the energy acceleration of the signal input to the communication unit 100 to determine whether the input signal is speech or noise based on the measurement. Communication unit. The voice activity detection mechanism of claim 1, wherein the voice activity detection mechanism includes a voice activity detection function 130 that performs detection every frame on input of signals to the voice activity detection mechanisms 130 and 135. Communication unit. The method of claim 2, wherein the detection per frame comprises the following frequency ranges: (i) full spectrum; (ii) spectral sub-bands; And, (iii) performing energy acceleration measurements on signal input to the voice activity detection mechanism (130, 135) for one or more ranges of spectral variance. 4. The voice activity detection mechanism of claim 3, wherein the voice activity detection mechanism comprises a voice activity determination function 135 operatively coupled to the voice activity detection function 130, such that the input signal is one of the measurements. A communication unit that determines whether it is negative based on the buffering operation of one or more measurements. 5. The communication unit according to claim 4, wherein said voice activity determining function (135) determines whether an input signal is negative using a frame average or rolling average of a plurality of said input signals. The communication unit according to claim 2, wherein if the energy acceleration measurement yields an energy acceleration value greater than the energy acceleration threshold, the input frame is estimated (265) as a negative frame. 7. The communication unit of claim 6, wherein the determination (265) that the input frame is a negative frame is applied retrospectively to a previous frame in a buffer of input signals. 8. A communication unit according to claim 6 or 7, wherein if the energy acceleration measurement yields an energy acceleration value that is greater than the energy acceleration threshold over a plurality of consecutive frames, the input signal is estimated (370) as a negative signal. The communication unit according to any one of claims 3 to 8, wherein if a sub-region of the input signal spectrum is selected, the selection is based on the sub-region most likely to include a basic pitch of the speech signal. The voice activity detection mechanism (130, 135) of any of the preceding claims, wherein the voice activity detection mechanism (130, 135) uses acceleration of voice-energy related features, eg, other voice or noise related metrics, eg. For example, the communication unit which validates parameter initialization of a spectrum subtraction method. A method of detecting a sound signal input to a communication unit, Measuring acceleration or change in energy of the input signal to the communication unit; And, Determining (315, 330, 350) whether the input signal is sound (370) or noise (375) based on the measuring step. 12. The method of claim 11, further comprising performing detection every frame on input of signals to the communication unit. 13. The method of claim 12, wherein in each frame the detection is: The following frequency ranges: (i) full spectrum; (ii) spectral sub-bands; And, (iii) performing an energy acceleration measurement on the input signal for one or more ranges of spectral variance, the sound signal input detection method. 14. A method according to any of claims 11 to 13, for determining whether a signal input to a communication unit is negative or noise. Determining (315) whether the input signal is negative (370) or noise (375) based on energy acceleration or change in energy measurement of the input signal using, for example, a frame average or rolling average of multiple input signals. 330, 350, further comprising determining whether the signal input to the communication unit is negative or noisy. The method of claim 14, wherein the determining step, If the energy acceleration measurement yields an energy acceleration value that is greater than an energy acceleration threshold, determining (265) that the input frame is a negative frame; And, Retroactively applying the determination to a previous frame in a buffer of input signals.