GB2430853A - Variable state change delay for a voice activity detector - Google Patents
Variable state change delay for a voice activity detector Download PDFInfo
- Publication number
- GB2430853A GB2430853A GB0519880A GB0519880A GB2430853A GB 2430853 A GB2430853 A GB 2430853A GB 0519880 A GB0519880 A GB 0519880A GB 0519880 A GB0519880 A GB 0519880A GB 2430853 A GB2430853 A GB 2430853A
- Authority
- GB
- United Kingdom
- Prior art keywords
- signal
- delay
- delay period
- speech
- value
- 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.)
- Granted
Links
- 230000008859 change Effects 0.000 title claims abstract description 60
- 230000000694 effects Effects 0.000 title claims abstract description 34
- 230000007704 transition Effects 0.000 claims abstract description 27
- 230000005236 sound signal Effects 0.000 claims description 9
- 238000000034 method Methods 0.000 claims description 8
- 238000004458 analytical method Methods 0.000 claims description 7
- 230000001934 delay Effects 0.000 claims description 7
- 230000003247 decreasing effect Effects 0.000 claims 1
- 230000001419 dependent effect Effects 0.000 abstract description 2
- 206010019133 Hangover Diseases 0.000 description 7
- 238000001514 detection method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 3
- 238000010295 mobile communication Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
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
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/78—Detection of presence or absence of voice signals
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04J—MULTIPLEX COMMUNICATION
- H04J3/00—Time-division multiplex systems
- H04J3/17—Time-division multiplex systems in which the transmission channel allotted to a first user may be taken away and re-allotted to a second user if the first user becomes inactive, e.g. TASI
- H04J3/175—Speech activity or inactivity detectors
Landscapes
- Engineering & Computer Science (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Mobile Radio Communication Systems (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
Usually, a Voice Activity Detector (VAD) 100 will deliver an output state indicating speech presence or absence. It is also known to introduce a constant delay period between a change of output states to confirm the presence (S) or absence of speech (NS) to reduce clipping of speech and to avoid falsely detecting speech. However, the constant delay does not produce the best results in all circumstances. The VAD described varies the length of the delay period dependent on the direction of output change (e.g. S to NS or vice versa) and a detected quality of the input signal. For example, if a S to NS transition is detected and the signal is poor, a longer delay period may be used to ensure the speech is not clipped. Conversely, if a S to NS transition with a good input signal quality is detected a shorter delay period can be used which is more efficient.
Description
1 2430853 TITLE: VOICE ACTIVITY DETECTOR
FIELD OF THE INVENTION
The present invention relates to a voice activity detector.
BACKGROUND OF THE INVENTION
A VAD (voice activity detector) is a device that analyses an input electronic signal representing an audio signal to determine whether or not speech is present. Usually, a VAD delivers an output signal that takes one of two possible values, respectively indicating that speech is detected to be present (S') or speech is not detected to be present (NS') . In general, the value of the output signal will change with time according to whether or not speech is detected to be present in the analysed signal.
A VAD is often incorporated in a speech communication device such as a fixed or mobile telephone, a radio or a like device. Use of a VAD is an important enabling technology for a variety of speech based applications such as speech recognition, speech encoding and hands free telephony. The primary function of a VAD is to provide an ongoing indication of speech presence as well to identify the beginning and end of a segment of speech. Devices such as automatic gain controllers employ a VAD to detect when they should operate in a speech present mode.
While VADs operate quite effectively in a relatively quiet environment, e. g. a conference room, they tend to be less accurate in noisy environments, such as in road vehicles, and in consequence they may generate detection errors. These detection errors include false alarms' which produce a signal indicating speech when none is present and mis-detects' which do not produce a signal to indicate speech when speech is present in noise.
There are many known types of algorithm employed in VADs to detect speech. A review of some different known types of VAD algorithms is given for example in Efficient voice activity detection algorithms using long-term speech information' in Speech Communication 42, (2004) pages 217-287. Each of the known algorithms has advantages and disadvantages. In consequence, some VADs may tend to produce false alarms and others may tend to produce mis-detects. Some VADs may tend to produce both false alarms and mis-detects in noisy environments.
In a VAD it is known to employ an analyser, which analyses a signal to determine whether or not speech is present, and in some cases, a state change delay which delays a change in state detected by the analyser (i.e. a state of speech being present or not present) from becoming effective until a delay period has elapsed. The delay period is known in the art as a hangover' and is typically one or more additional speech frames analysed.
This delay period is applied, for example, to reduce the clipping of speech at the end of a speech segment (where speech energy often falls to zero over several frames) and to avoid false alarms due to transient noise blips which might be wrongly detected as the beginning of speech following a period of no speech. In the prior art the delay period or hangover applied by the state change delay is a constant delay period for each given transition type (speech S' to no speech NS' or NS' to S') . Typically, the delay period is set at about 7 or 8 frames for an S' to NS' transition and about 1 to 3 frames for an NS' to S' transition. The present inventors have found that, unfortunately, a constant delay period as typically selected and applied in the
prior art is not ideal in all circumstances.
StTh(ARY OF THE INVENTION According to the present invention in a first aspect there is provided a voice activity detector as defined in claim 1 of the accompanying claims.
Further features of the present invention are as defined in the accompanying dependent claims and are disclosed in the embodiments of the invention to be described.
The present invention provides a voice activity detector which includes an analyser for analysing an input signal representing an audio signal to determine if speech is present in the represented audio signal and for producing an output state signal indicating whether or not speech is detected by the analyser and at least one state change delay which applies a delay period (hangover') which has a variable value and is set adaptively according to a value of current signal quality indicated by an output signal from a signal quality estimator. The signal quality estimator enables a current measure of analysed signal quality to be obtained. The signal quality estimator may estimate current signal to noise ratio or other known signal quality parameter.
Beneficially, the delay period applied by the state change delay may be adaptively set to a value which suits the current circumstances, i.e. whether or not the signal quality is good and on the type of transition (state change) detected, e.g. either from S' to NS' or from NS' to S' For example, where an S' to NS' transition is detected and the signal quality is relatively poor, a relatively long period may be used. The delay period applied for such a transition in the prior art with poor signal quality may be too short to avoid speech clipping. On the other hand, where an S' to NS' transition is detected and the signal quality is relatively good, a relatively short period may beneficially be used. The delay period applied for such a transition in the prior art may unnecessarily be too long and this can lead to a waste of valuable energy.
For example, where the VAD is used in a wireless communications device in conjunction with a transmitter or receiver which is switched on only when speech is detected, valuable battery energy is wasted if the transmitter or receiver is switched on for longer than necessary. In addition, it is beneficial to minimise the time for which a transmitter using the VAD is switched on in order to reduce over the air interference with other wireless signals. Thus, by avoiding an unnecessarily long delay period for an S' to NS' transition in the VAD of the invention, energy may be saved and interference may be minimised by use of a shortened delay period where the signal quality is detected to be relatively good.
Where an NS' to S' transition is detected and the detected signal quality is relatively poor, a relatively short delay period may be used in the VAD of the invention. This beneficially allows the start of speech to be detected quickly in such conditions. The fixed delay period applied in the prior art may unnecessarily be too long causing clipping at the start of a speech segment. Where an NS' to S' transition is detected and the detected signal quality is relatively high, the delay period may be increased with reduced risk of not detecting the start of speech.
Conveniently, the delay period adaptively selected respectively for each of an S' to NS' transition and for a NS' to S' transition is integral to a number of speech frames analysed.
Detailed examples of the ways in which the delay period may be adaptively selected according to current circumstances detected will be described in the embodiments of the invention which follow.
Embodiments of the present invention will now be described by way of example with reference to the accompanying drawings, in which:
BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWINGS
FIG. 1 is a block schematic diagram showing a voice activity detector embodying the present invention.
FIG. 2 is a block schematic diagram of a modified voice activity detector embodying the present invention.
DESCRIPTION OF EMBODINTS OF THE INVENTION
In this description the expressions speech
segments' and speech frames' are used. Speech segments are the parts of a signal where speech is present. In the parts where no speech is present, the signal may be considered to consist of noise segments. Typically, a speech segment lasts for one to two seconds. Frames are the consecutive units into which a signal is divided in order to be analysed to determine whether speech is present in the unit of the signal being analysed.
Typically, the length of a frame is from 20 msec to 30 msec depending on the particular analysis procedure to be used.
FIG. 1 is a block schematic diagram showing a VAD (voice activity detector) 100 embodying the present invention. An input electrical signal Ii to be analysed, representing an audio signal, is applied to an analyser 101 which operates a voice activity sensing algorithm to analyse frames of the input signal to determine if speech is present or not in each frame. The algorithm operated by the analyser 101 may be known per se. For example, the algorithm may be one of: (i) a pitch detection algorithm; (ii) an LPC (linear prediction coding coefficients) distance algorithm; and (iii) a cepstral distance algorithm. In an embodiment of the invention, the analyser 101 uses a cepstral distance algorithm. The cepstral distance algorithm uses cepstral analysis' which is a well known non-linear technique to detect perceptual harmonic cepstral' coefficients for speech recognition in a given signal.
Using a form of this algorithm, the input signal Ii is divided into 20 milliseconds frames. For each frame, cepstral coefficients are calculated. Then, the analyzer N 2 101 computes a parameter (cep(i)-cep_noise(i)) , where cep(i) is the i-th cepstral coefficient of the current frame, and cep_noise(i) is the i-th cepstral coefficient of the noise. Further, in the above computation, N is the order of associated linear prediction coefficients. The analyzer 101 calculates cep_noise(i) according to an internal decision from the VAD that no speech is present, and updates the calculation with a factor obtained from a first order hR (infinite impulse response filter) . If the result is below a set threshold then the algorithm in the analyzer 101 gives a no speech' indication; otherwise speech' is determined to be present.
The analyser 101 produces an output state signal 01 having one of two possible values, namely a value S' (speech detected) or NS' (speech not detected) . The value of the output signal 01 can change with time between frames if a transition between speech and no speech segments is detected. The output state signal 01 is applied to a state change delay 103. The state change delay 103 detects whether there is a change of value, i.e. indicating an S' to NS' transition or an NS' to S' transition as appropriate. Where such a change is detected, the state change delay 103 maintains the present value of the output state signal 02 for a delay period (hangover') consisting of one or more additional frames calculated in a manner to be described later. An output state signal 02 is produced by the state change delay 103. If the change of value of the state signal 01 is still indicated to be correct after the applied delay period, the value of the state signal 02 is changed to match the changed value of the state signal 01. Thus, the state change delay 103 produces an output state signal 02 which has the same value as the output state signal 01 from the analyser 101 except that a change in the value of the state signal 01 is allowed in the state of the output state signal 02 only if the change is determined to be correct after the given delay period.
In order for such a change to be determined to be correct, a value of the state signal 01 which has changed to a new value must be constant and equal to the new value for all frames of the delay period, otherwise the value of the output state signal 02 is unchanged (from the value immediately prior to the applied delay period) The VAD 100 includes a signal quality estimator 105 and a delay calculator 113. The signal quality estimator includes a signal power estimator 109 and a noise power estimator 111. The signal power estimator 109 and the noise power estimator 111 receive the input signal Ii. The signal power estimator 109 estimates in a known manner a power value for a current frame of the received signal Ii. The current frame is the same as that analysed by the analyser 101. The noise power estimator 111 estimates a noise value in a known manner. The noise power estimator 111 may for example use a first order uP, filter as referred to earlier to carry out an estimation of the detected received power for the current frame in a known manner. Preferably, the noise power estimation is not carried out in every frame, but only in frames that are determined by the analyser 101 not to contain speech.
Signals representing the values estimated by the signal power estimator 109 and the noise power estimator 111 are delivered to a divider 112 which determines a value of the signal power value divided by the noise power value. The divider 112 thereby produces an output signal 03 representing a signal to noise ratio as a signal quality value for the current frame. The signal 03 is delivered to and stored by the delay calculator 113 and used by the delay calculator 113 in one of the ways described later. An output signal 04 is produced by the delay calculator 113 and is delivered to the state change delay 103. When a transition is detected by the state change delay 103, the delay period applied by the state change delay 103 is the latest value of delay value which the state change delay 103 has received from the delay calculator 113 as indicated by the signal 04.
The VAD 100 includes a current state detector 107 which receives either the output state signal 01 produced by the analyser 101 or the output state signal 02 produced by the state change delay 103. The current state detector 107 thereby determines for the frame immediately before that currently being analysed by the analyser 101 whether a segment of speech or a segment of noise is present. The type of segment detected indicates the current state of the VAD 100. The current state detector 107 applies a current state signal indicating the detected current state to the delay calculator 113 accordingly. When the value of the current state signal applied by the current state detector 107 to the delay calculator 113 indicates speech present (S'), i.e. that the next transition will be S' to NS', the delay calculator 113 is activated to calculate a first delay period by a first procedure using the signal 03 provided by the divider 112, e.g. in a manner to be described later. When the value of the current state signal applied by the current state detector 107 to the delay calculator 113 indicates speech not present (NS'), i.e. that the next transition will be NS' to S', the delay calculator 113 is activated to calculate a second delay period by a second procedure using the signal 03 provided by the divider 112, e.g. in a manner to be described later. The first and second delay periods calculated by the delay calculator 113 can vary as the value of the signal 03 from the divider 112 changes.
Thus, each of the first and second delay periods are set adaptively, according to the measured signal quality as indicated by the signal 03, e. g. in a manner to be described later. The first delay period may for example be in the range of from 2 frames to 30 frames, more particularly from 5 frames to 15 frames, and the second delay period may for example be in the range of from 1 frame to 5 frames.
The delay calculator 113, the state change delay 103 and the current state detector 107 may be incorporated within a common delay unit 117. In practice, the common delay unit 117 and all other components of the VAD 100 may be fabricated in the form of an integrated circuit such as a semiconductor microprocessor programmed to carry out the functions of the VAD 100.
The delay calculator 113 operates an algorithm to calculate a delay period or hangover' to be applied currently by the state change delay 103 by employing a current value of the signal 03 which it receives from the signal quality estimator 105. Examples of how the delay period may be calculated by the delay calculator 113 are described as follows.
Where the current state detector 107 detects speech currently present and indicates this in the current state signal delivered to the delay calculator 113, the delay calculator 113 may use the current signal quality value indicated by the signal 03 as follows. The delay calculator 113 may calculate a first delay period which is an inverse function of the signal quality. In other words, when the signal 03 indicates a poor (low) signal quality, the first delay period is relatively high, and when the signal 03 indicates a relatively good (high) signal quality, the delay period is relatively low. As noted earlier, the first delay period applied may for example be in a range between a minimum delay period of 2 frames and a maximum delay period of 30 frames, more particularly between 5 and 15 frames. Each value of the first delay period is conveniently selected to be an integral number of frames.
Where speech is currently detected by the current state detector 107 and this is indicated in the current state signal delivered to the delay calculator 113, the delay calculator 113 may use the received values of the signal 03 for consecutive frames to calculate a signal quality value averaged over a number of frames. The delay calculator 113 may use the received signal 03 to calculate a first average signal quality value averaged over all frames for the current detected speech segment up to the frame immediately preceding that currently being analysed by the analyser 101. Alternatively, or in addition, the delay calculator 113 may calculate a second average signal quality value averaged over a finite (fixed) number of frames, e.g. ten frames, up to the frame immediately preceding that currently being analysed by the analyser 101. Alternatively, or in addition, the delay calculator 113 may calculate for the current detected speech segment, for frames up to the frame immediately preceding that currently being analysed by the analyser 101, a maximum value of the first average signal quality value and/or a maximum value of the second average signal quality value.
One particular way in which the delay calculator 113 may suitably employ the first average signal quality value and the second average signal quality value is for example to calculate a first delay period Dl according to a relationship: Dl = a +(b/Qs)- cQm Equation 1 where a, b and c are predetermined constants, Q3 is an average value of the first signal quality value and Qm is a maximum value for the current detected speech segment of the second signal quality value. The first delay period Dl is then conveniently chosen as an integral nurttber of frames by rounding the result of Dl from Equation 1 to the next highest integer.
Where the current state detector 107 detects that speech is currently not present and provides a current state output signal to the delay calculator 113 accordingly, the delay calculator 113 may for example use the current value of the signal 03 in a particular form of the second procedure as follows. The delay calculator 113 may calculate a second delay period D2 which is an increasing function of the signal quality value indicated by the signal 03. In other words, when the value of the signal 03 is relatively low, the second delay period D2 is set to be low, and when the value of the signal 03 is higher, the second delay period D2 is set to be higher. In practice, because noise is being detected, even a relatively high' signal quality value obtained will be low when compared with the much higher values obtained when speech is present.
The second delay period which is selected by the delay calculator 113 may, as noted earlier, be in a range between a minimum delay period of one frame and a maximum of five frames. The second delay period is conveniently selected to be an integral number of frames.
Thus, the delay period (hangover') calculated for each frame by the delay calculator 113 is provided to the state change delay 103 by the signal 04 and is employed by the state change delay 103 on a frame by frame basis to update the delay period to be applied to delay any detected change in the current value of the signal 01 to be allowed in the output signal 02 until the applied delay period has elapsed.
We have compared the VAD 100 embodying the invention with a prior art voice activity detector using the same VAD analyser (analyser 101 in the VAD 100) and a known signal as the analysed input signal representing an audio signal. The prior art detector used a fixed delay period of seven frames for an S' to NS' transition and two frames for a NS' to S' transition.
The results obtained showed better performance by the VAD 100 embodying the invention. The results showed that there is much less clipping of speech segments where the signal quality is low, and, where the signal quality is suitably high, the delay periods may be shortened without affecting performance. As noted earlier, using such shortened delay periods when the measured signal quality is suitably high, beneficially allows less processing power to be used (e.g. by reducing a period for which an associated transmitter or receiver is switched on) and this leads to a saving in battery life.
FIG. 2 is a block schematic diagram of a modified VAD 200 embodying the present invention. The modification of this embodiment is an example of the use of multiple VADs to produce a single decision as to whether or not speech is present. Such an arrangement and a procedure using it is the subject of a copending UK patent application No. GB0518213.4 by the present Applicant. In the VAD 200, an input signal is applied in parallel to a first analyser 201, a second analyser 205 and a third analyser 209. Each of the first analyser 201, the second analyser 205 and the third analyser 209 operates a voice activity sensing algorithm to analyse frames of the input signal to determine if speech is present or not. The algorithms operated are different in each case although each of the algorithms may be known per se. For example, the different algorithms may be the particular known algorithms referred to earlier. Output state signals are produced by each of the analysers 201, 205 and 209. An output state signal 05 is produced by the analyser 201 and is applied to a state change delay 203. An output state signal 06 is produced by the analyser 205 and is applied to a state change delay 207.
An output state signal 07 is produced by the analyser 209 and is applied to a state change delay 211.
The state change delay 203 produces an output state signal 08. The state change delay 203 detects whether there is a change of value, i.e. indicating an S' to NS' transition or an NS' to S' transition as appropriate in the state signal 05 which it receives from the analyser 201. Where such a change is detected, the state change delay 203 maintains the present value of the output state signal 08 for a delay period (hangover') consisting of one or more additional frames. If the change of value of the state signal 05 is still indicated to be correct after the applied delay period, the value of the state signal 08 is changed to match the changed value of the state signal 05. The state change delays 207 and 211 operate in a manner similar to the state change delay 203 producing output state signals 09 and 010 respectively.
The VAD 200 includes a combiner 213 which receives as input signals the output state signals 08, 09 and 010 from the state change delays 203, 207 and 211 respectively. The combiner 213 applies a combining algorithm to the signals 08 to 010 to produce a single output state signal Oil. The combiner 213 may operate in one of the ways described by the Applicant in the copending UK patent application referred to earlier. The output state signal Oil is applied to a further state change delay 215 which operates in a manner similar to the state change delay 203 producing an overall output state signal 012.
In accordance with an embodiment of the invention, one or more, preferably all, of the state change delays 203, 207, 211 and 215 operate to apply a delay which is adaptively set in the manner described earlier with reference to FIG. 1 each using its own delay calculator (operating in a manner similar to the delay calculator 113 described earlier with reference to FIG. 1) based on an output signal from a signal quality estimator (operating in a manner similar to the signal quality estimator 105 described earlier with reference to FIG. 1).
The VADs 100 and 200 embodying the invention which have been described with reference to FIGS. 1 and 2 may be used in any of the known applications in which voice activity detectors are used. Thus, the VADs 100 and 200 may for example be used in landline telephony applications and in mobile communications, particularly in transmitters or receivers of portable and/or mobile communications terminals. The VAD 100 or 200 may for example be used in conjunction with a transmitter or a receiver to determine when to switch the transmitter or receiver or a component thereof, e.g. an automatic gain control (AGC) circuit of a receiver, on and off.
Claims (18)
1. A voice activity detector including an analyser for analysing a input signal representing an audio signal to determine if speech is present in the represented audio signal and for producing an output state signal accordingly; a state change delay for receiving an output state signal from the analyser, for detecting a change in the output state signal from the analyser and for providing an output state signal in which, following a detected change in the output state signal from the analyser, a current state is maintained until an applied delay period has elapsed; and a signal quality estimator for estimating a quality value of the input signal and for producing an output signal indicating the estimated value; wherein the state change delay is operable to apply a delay period which has a variable value which is determined adaptively according to a current value of the output signal produced by the signal quality estimator.
2. A voice activity detector according to claim 1 wherein the signal quality estimator is operable to estimate for a current frame of the input signal a value of total power and a value of noise power and to divide the value of total power by the value of noise power to provide an estimate of signal to noise ratio for the current frame.
3. A voice activity detector according to claim 1 or claim 2 including a delay calculator operable to receive an output signal from the signal quality estimator and to employ the received signal to calculate a delay period to be applied for a current frame by the state change delay.
4. A voice activity detector according to any one of the preceding claims wherein the state change delay is operable to apply a first adaptively set delay period when a transition from speech present to speech not present is detected and to apply a second adaptively set delay period when a transition from speech not present to speech present is detected.
5. A voice activity detector according to claim 4 wherein the first adaptively set delay period is longer than the second adaptively set delay period.
6. A voice activity detector according to claim 4 or claim 5 wherein the first adaptively set delay period is a period in the range of from 2 frames of the input signal to 30 frames of the input signal.
7. A voice activity detector according to any one of claims 4 to 6 wherein the second adaptively set delay period is a period in the range of from 1 frame of the input signal to 5 frames of the input signal.
8. A voice activity detector according to any one of claims 4 to 7 including a delay calculator operable to receive an output signal from the signal quality estimator and to employ the received signal to calculate a delay period to be applied by the state change delay and wherein the delay calculator is operable to employ a first relationship between delay period and signal quality to calculate the first delay period and to employ a second relationship between delay period and signal quality to calculate the second delay period.
9. A voice activity detector according to claim 8 wherein the first relationship comprises the first delay period being a function of the signal quality value indicated by the output signal from the signal quality estimator wherein the delay period is decreased with an increase in the signal quality value.
10. A voice activity detector according to claim 8 or claim 9 wherein the second relationship comprises the second delay period being a function of the value of the signal quality indicated by the output signal from the signal quality estimator wherein the delay period is increased with an increase in the signal quality value.
11. A voice activity detector according to any one of the preceding claims B to 10 including a current state detector operable to receive an output state signal provided by the analyser or by the state change delay for a preceding frame of the input signal and, depending on the received output state signal, to provide a current state signal to the delay calculator, wherein the delay calculator is operable to calculate a delay period by a first procedure when the current state signal indicates speech present and to calculate a delay period by a second procedure when the current state signal indicates speech not present.
12. A voice activity detector according to any one of claims 3 to 12 wherein the delay calculator is operable to calculate a delay period using a relationship which is a function of a signal quality for a current detected speech segment and of a signal quality for a finite number of frames of the input signal.
13. A voice activity detector according to claim 12 wherein the delay calculator is operable to calculate a delay period Dl when speech is currently detected to be present according to a relationship: Dl = a + (b/Qs)- cQm where a, b and c are predetermined constants, Q5 is an average value of signal quality for frames since a detected beginning of a current speech segment and Qm is a maximum value of a current speech segment of an average signal quality taken over a finite of frames of the input signal.
14. A voice activity detector according to claim 13 wherein Qm is a maximum value of average signal quality of a current speech segment taken over a finite fixed number of frames of the input signal ending with the frame immediately preceding that currently analysed.
15. A voice activity detector according to any one of the preceding claims wherein the state change delay is operable to apply a delay period which is an integral number of frames.
16. A voice activity detector according to any one of the preceding claims including a plurality of analysers each operable to analyse an input signal representing an audio signal to determine if speech is present in the represented audio signal and for producing an output state signal accordingly, a plurality of state change delays each being operably coupled to an associated one of the analysers and each being operable to produce an output state signal, and a combiner for producing a combined state value using output state signals produced by each of the state change delays, wherein each of the state change delays is operable to apply a value of a delay period which is variable and determined adaptively according to an estimated signal quality of the analysed signal.
17. A voice activity detector according to claim 16 including an output state change delay operably coupled to the combiner to receive an output signal comprising an output state value produced by the combiner and operable to apply a value of a delay period which is variable and determined adaptively according to an estimated signal quality of the analysed signal.
18. A voice activity detector according to any one of the preceding claims wherein the voice activity detector is in the form of an integrated circuit.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
GB0519880A GB2430853B (en) | 2005-09-30 | 2005-09-30 | Voice activity detector |
PCT/US2006/034350 WO2007040883A2 (en) | 2005-09-30 | 2006-09-02 | Voice activity detector |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
GB0519880A GB2430853B (en) | 2005-09-30 | 2005-09-30 | Voice activity detector |
Publications (3)
Publication Number | Publication Date |
---|---|
GB0519880D0 GB0519880D0 (en) | 2005-11-09 |
GB2430853A true GB2430853A (en) | 2007-04-04 |
GB2430853B GB2430853B (en) | 2007-12-27 |
Family
ID=35395002
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
GB0519880A Expired - Fee Related GB2430853B (en) | 2005-09-30 | 2005-09-30 | Voice activity detector |
Country Status (2)
Country | Link |
---|---|
GB (1) | GB2430853B (en) |
WO (1) | WO2007040883A2 (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB2352378A (en) * | 1999-05-28 | 2001-01-24 | Mitel Corp | Detecting silence in voice packets |
US20030198328A1 (en) * | 2002-04-17 | 2003-10-23 | Dunling Li | Voice activity identification for speaker tracking in a packed based conferencing system with distributed processing |
US20050055201A1 (en) * | 2003-09-10 | 2005-03-10 | Microsoft Corporation, Corporation In The State Of Washington | System and method for real-time detection and preservation of speech onset in a signal |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2001033814A1 (en) * | 1999-11-03 | 2001-05-10 | Tellabs Operations, Inc. | Integrated voice processing system for packet networks |
-
2005
- 2005-09-30 GB GB0519880A patent/GB2430853B/en not_active Expired - Fee Related
-
2006
- 2006-09-02 WO PCT/US2006/034350 patent/WO2007040883A2/en active Application Filing
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB2352378A (en) * | 1999-05-28 | 2001-01-24 | Mitel Corp | Detecting silence in voice packets |
US20030198328A1 (en) * | 2002-04-17 | 2003-10-23 | Dunling Li | Voice activity identification for speaker tracking in a packed based conferencing system with distributed processing |
US20050055201A1 (en) * | 2003-09-10 | 2005-03-10 | Microsoft Corporation, Corporation In The State Of Washington | System and method for real-time detection and preservation of speech onset in a signal |
Also Published As
Publication number | Publication date |
---|---|
WO2007040883A3 (en) | 2007-06-21 |
GB2430853B (en) | 2007-12-27 |
GB0519880D0 (en) | 2005-11-09 |
WO2007040883A2 (en) | 2007-04-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US8977556B2 (en) | Voice detector and a method for suppressing sub-bands in a voice detector | |
CN101320559B (en) | Sound activation detection apparatus and method | |
US9390729B2 (en) | Method and apparatus for performing voice activity detection | |
US5963901A (en) | Method and device for voice activity detection and a communication device | |
US6360199B1 (en) | Speech coding rate selector and speech coding apparatus | |
KR100883712B1 (en) | Method of estimating sound arrival direction, and sound arrival direction estimating apparatus | |
EP3726530B1 (en) | Method and apparatus for adaptively detecting a voice activity in an input audio signal | |
US20120232896A1 (en) | Method and an apparatus for voice activity detection | |
JP2573352B2 (en) | Voice detection device | |
RU2720357C2 (en) | Method for estimating background noise, a unit for estimating background noise and a computer-readable medium | |
GB2384670A (en) | Voice activity detector and validator for noisy environments | |
US20120265526A1 (en) | Apparatus and method for voice activity detection | |
US20060265219A1 (en) | Noise level estimation method and device thereof | |
US8442817B2 (en) | Apparatus and method for voice activity detection | |
US6842526B2 (en) | Adaptive noise level estimator | |
US20050171769A1 (en) | Apparatus and method for voice activity detection | |
CN100492495C (en) | Apparatus and method for detecting noise | |
EP1561298A1 (en) | Combining direct interference estimation and decoder metrics for amr mode adaptation in gsm systems | |
GB2430853A (en) | Variable state change delay for a voice activity detector | |
JP4352875B2 (en) | Voice interval detector | |
JP2002300687A (en) | Electronic apparatus | |
GB2430129A (en) | Voice activity detector | |
JPH09113350A (en) | Method and apparatus for predicting average level of background noise | |
JPH07212296A (en) | Vox control communication equipment | |
EP1551006A1 (en) | Apparatus and method for voice activity detection |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
732E | Amendments to the register in respect of changes of name or changes affecting rights (sect. 32/1977) |
Free format text: REGISTERED BETWEEN 20120419 AND 20120425 |
|
PCNP | Patent ceased through non-payment of renewal fee |
Effective date: 20120930 |