KR20060027570A - Voice dialing apparatus and method using features of the voice - Google Patents

Abstract

The present invention discloses a voice dialing device and method using the features of voice. The apparatus for dialing by voice without pressing a button when dialing a mobile communication terminal includes an input interface for extracting voice feature parameters from frame-by-frame packet data output from a vocoder, and a voice for detecting voice segments using voice feature parameters. A section detection section, a feature extraction section for extracting at least one of the number of peaks and peak levels of the speech signal in the detected speech section as features of the speech recognition speech signal, and a feature of the speech recognition speech signal to each telephone number. A voice register that registers in the database so as to correspond to the voice register; and a voice recognizer that selects the closest voice signal by comparing the voice signal registered in the database with the voice signal registered in the database using the features of the voice signal for voice recognition; Output voice signal and phone number corresponding to voice recognition result It is characterized by comprising a result output.

Description

Voice dialing apparatus and method using features of the voice}

1 is a block diagram of a voice dialing apparatus using the features of voice according to the present invention.

FIG. 2 is a block diagram of the voice section detector 120 shown in FIG. 1.

3 is an exemplary block diagram of the voice register 140 shown in FIG.

4 is a block diagram of an embodiment of the present invention of the speech recognition unit 150 shown in FIG.

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a dialing technique using voice in a mobile terminal, and more particularly, to an apparatus and method for voice dialing using a feature of real-time implementation using only software without adding hardware to an existing terminal using feature parameters of each vocoder rather than a voice signal It is about.

Voice dialing technology is an application of voice recognition technology to phone control. This voice dialing technology used 8KHz sampling, 16-bit PCM data or equivalent voice directly as input. However, to implement voice dialing techniques in software alone without the addition of hardware under a mobile phone environment, one must find a way to use the mobile phone's vocoder results. This is because, in order to receive voice directly, a large amount of memory is required as well as an A / D device.

In addition, the method of reproducing speech using the vocoder result is generally insufficient in the amount of computation to be implemented by software alone. On the other hand, when voice dialing in a hands-free environment, another problem arises, which is background noise processing. In other words, to make a call from a car running on the highway, a hands-free device should be used. In this case, the microphone is usually mounted on a car sun visor. In this environment, voice is generated, and voice and noise generated on the highway are simultaneously introduced. When measuring highway noise in a car that is running more than 100 km, the signal-to-noise ratio (SNR) ranges from -6 dB to 6 dB. This value means that the output size of voice or noise is similar. Therefore, there is a problem that voice recognition should be performed in a variety of high-value noise environments with the problem of using packet data, which is a vocoder output rather than voice.

There are many known techniques to remove noise for speech recognition in noisy environments, such as Spectral Substraction, but this method assumes that PCM data is available.

On the other hand, speech recognition using gain-related parameters of the vocoder produces up to 90% recognition rate in low noise environment, but only 40% or less recognition rate in noisy environment. Cannot be guaranteed.

The technical problem to be achieved by the present invention is the voice using the voice feature that can apply the voice dialing function to the existing mobile phone using the voice features such as the number of peaks and / or the peak level of the voice to the existing mobile phone only without the addition of hardware The present invention provides a dialing device and method.

In order to solve the above technical problem, the voice dialing device using the voice feature according to the present invention for dialing by voice without pressing a button during dialing of a mobile communication terminal includes a voice feature parameter from the packet data for each frame output from a vocoder. A speech section detector for detecting a speech section by using an input interface to be extracted, the speech feature parameter output from the input interface, and at least one of the number of peaks of the speech signal and the level of the peaks in the detected speech section A feature extraction unit for extracting a feature of the voice signal for speech recognition as a feature; a voice register for registering a feature of the voice recognition voice signal extracted from the feature extractor into a database so as to correspond to each telephone number; The speech signal for speech recognition extracted from the A voice recognition unit that selects the closest voice signal by comparing the voice signal registered in the database with the voice spoken for voice dialing by using the feature, and outputs a voice signal and a telephone number corresponding to the voice recognition result It is preferable that it consists of an output part.

In order to solve the above other technical problem, the voice dialing method using the feature of the voice dialing to the voice when dialing the mobile communication terminal, converting the packet stream signal output from the vocoder into an unpacking parameter stream signal Detecting a section of the unpacking parameter stream signal, detecting at least one of the number of peaks of the audio signal present in the detected signal and the level of the peak as a feature of the audio signal; A voice registration step of storing a feature of the signal in a memory so as to correspond to each telephone number, a voice recognition step of selecting a voice signal closest to the voice spoken from the registered voice signal, a voice signal corresponding to the voice recognition result, and Preferably, the step comprises outputting a telephone number.

Hereinafter, a voice dialing apparatus using a voice feature according to the present invention will be described with reference to the accompanying drawings.

1 is a block diagram of a voice dialing apparatus using features of a voice according to the present invention, which includes an input interface 110, a voice section detector 120, a feature extractor 130, a voice register 140, and a voice recognizer ( 150 and the result output unit 160.

The input interface 110 unpacks the packet data of the vocoder, interprets the data, and inputs the data to the voice interval detector 120. It extracts feature parameters within 1 ~ 2msec from packet data per frame.

The voice section detector 120 detects the voice section by using a gain characteristic of a parameter generated by the input interface 110.

The feature extractor 130 extracts at least one of the number of peaks and the level of peaks of the speech signal belonging to the speech section detected by the speech section detector 120 as the feature of the speech recognition voice signal. Here, the number of peaks of the speech signal belonging to the speech section varies depending on the telephone number. In addition, the level of the peak of the voice signal belonging to the voice section also varies depending on the telephone number. In this case, the feature extractor 130 may extract only the number of peaks of the voice signal belonging to the voice section as the feature of the voice signal, or may extract the level of the peak of the voice signal as the feature of the voice signal. Alternatively, the feature extractor 130 may extract both the number of peaks of the speech signal and the level of the peak of the speech signal as the features of the speech signal.

On the other hand, the voice register 140 stores the feature of the voice signal corresponding to each telephone number in a memory (Flash ROM).

The speech recognition unit 150 calculates how similar to the registration database 170 in which the current speech is stored. Recognition is performed once or if necessary with two voices.

The result output unit 160 performs a job of confirming to the user the voice corresponding to the recognition result before making a call.

The operation of the present invention will be described based on the above configuration.

The present invention relates to voice dialing for a mobile phone, and does not receive a voice but extracts a feature of a voice signal for speech recognition with an arbitrary codec packet value and detects a voice to enable an interface between a human and a machine.

The present invention uses only speech feature parameters generated in vocoder such as QCELP, EVRC and RPE-LTP. The input interface 110 unpacks the packet data of the vocoder, interprets the data, inputs the data to the voice interval detector 120, and extracts feature parameters within 1 to 2 msec from the packet data for each frame.

FIG. 2 is a block diagram of the voice section detector 120 of FIG. 1. The input interface unit 210, the similar signal generator 220, the similar gain generator 230, the gain average generator 240, And a frame state determination unit 250 and a post processing unit 250.

The similar signal generator 220 generates a similar signal using the feature parameter. To this end, the pseudo signal generator 220 performs gain filtering using a vocoder gain value on a white noise pulse, and then performs pitch filtering using pitch information. The signal generated in this way is not a speech signal that can be distinguished by a human ear, but contains loudness and frequency information, which is basic information for distinguishing speech. Equation 1 below generates an analogous signal.

x (i) = w (i) * G + x (i-l) * B

Where i is a time variable corresponding to the number of samples, G is a gain per frame, L is a pitch per frame, B is a pitch gain per frame, x (i) is a similar signal, and w (i) is a white signal. Represent each.

Using this method, the signal reproducing device of each vocoder can be replaced for real-time implementation of the voice detector in the mobile phone.

The similar gain generator 230 is a number obtained by taking an absolute value of the generated similar signal and adding the frame size. Each signal must be squared, summed, and square rooted to produce an accurate gain, but multiplication or square root cannot be used for real-time implementation. Equation 2 is an equation for generating the gain of the pseudo signal.

s (j) = abs_sum (x (j * I) to x (j * (I + 1)-1))

Here, j corresponds to the number of frames j = i / I. I is a frame size of 160 (20 msec) in general. abs_sum (x (k) to x (m)) means the sum of absolute values from the kth similar signal to the mth similar signal.

The gain average generator 240 calculates an average value of four consecutive frames as shown in Equation 3 below. The reason for using four consecutive frames in the present invention is that the feature of speech is the length of the frame that is best maintained.

nG (j) = (s (j-3) + s (j-2) + s (j-1) + s (j)) / 4

The gain of the similar signal for each frame has a large variation in time, and thus the probability of error occurring when detecting a voice using the result of Equation 2 increases. Equation 3 has a more stable variation in gain than Equation 2, so that it is easy to distinguish between the voice part and the noise part.

The frame state determination unit 250 determines whether the current frame is a voice, a silent section between the voice and the voice, or a background noise section using the gain average value of the similar signal and the past state determination result. The first speech segment determination informs the start position and the end position information including the silence section between the speech and the speech.

The post processor 260 corrects the start position and the end position in more detail using the first speech section result. This performs the state determination step again within 10 frames before and after the start position result and 10 frames before and after the end position result to correct the voice detection result.

Therefore, the voice section detected by the voice section detector 120 is used by the feature extractor 130 to extract the feature of the voice signal for voice recognition. Then, the feature of the voice recognition voice signal of the feature extraction unit 130 is input to the voice registration unit 140 in the registration mode, the voice recognition unit 150 in the recognition mode.

The voice register 140 stores the feature of the voice signal in a memory (Flash ROM) so as to correspond to each telephone number. Also, the voice feature is extracted using the result of analyzing the vocoder's packet.

FIG. 3 is an exemplary block diagram of the voice register 140 shown in FIG. 1 and includes a registration database comparator 310, a speech comparator 320, and a database storage 330.

The registration database comparison unit 310 shown in FIG. 3 compares the similarity between the input voice and the voice registered in the database 170. The speech comparison unit 320 compares the speech characteristics of the speech sound with the resounding sound when the speech sound is not similar to the voice registered in the database. The database storage unit 330 stores the speech characteristics of the speech sound when the speech characteristics compared in the speech comparison unit 320 are the same. In addition, the speech comparison unit 320 may further include an additional speech comparison unit (not shown) for adding the input voice to re-compare the speech characteristics when the speech characteristics of the speech sound and the re-uttered speech sound are different.

Voice registration requires two to three voice utterances. Referring to the registration process, when the voice sound 1 is inputted, the voice sounder 1 is stored in the voice register 140 when it is determined to be similar to the previously stored voice databases.

When the speech sound 2 is input, the similarity between the speech sound 1 and the speech sound 2 is examined, and if it is determined that the speech sound 2 is similar, the characteristics of the speech sound 1 and the speech sound 2 are stored in the voice register 140. If the sound 1 and the sound 2 are not similar, input the sound 3 to examine the similarity between the sound 1 and the sound 2 to the sound 3, and if the sound 1 or the sound 2 and the sound 3 are similar. The voice register 140 is stored.

Here, the additional voice comparison unit (not shown) refers to a step of comparing the voice 3 and the voice 1 or the voice 2. The registration database is stored in flash ROM so that it can be written and read at all times.

The speech recognition unit 150 calculates how similar to the registered database in which the current speech is stored. Recognition is performed once or if necessary with two voices.

4 is a block diagram of an embodiment of the voice recognition unit 150 shown in FIG. 1 according to the present invention, and includes a registration database comparison unit 410 and a recognition confirmation unit 420.

The registration database comparison unit 410 shown in FIG. 4 compares the similarity between the input voice and the voice registered in the database. In addition, the registration database comparator 410 adds an input voice to re-compare the voice registered in the database if the input voice is not similar to the voice registered in the database (not shown). Can be added further. The recognition check unit 420 determines whether to make a call with the recognition candidate determined by the registration database comparison unit 410. Even if the recognition result is the top candidate, don't call if the result is unreliable. To do this, we use the current noise level, the difference in recognition results between the first and second candidates, and the similarity values between the candidates.

The voice registration unit 140 or the voice recognition unit 150 extracts the feature using the result of analyzing the vocoder packet.

For example, a CDMA terminal uses the quantized LSP (Line Spectrum Pair) coefficient to create a recognition characteristic called pseudo-string, and a GSM terminal uses a LAR. Create a feature for recognition using the Log Araea Ratio coefficient.

The result output unit 160 performs a job of confirming and hearing the voice corresponding to the recognition result before the call is made. To do this, only the vocal parts are stored in the memory together with the recognition feature in the voice register.

Voice dialing apparatus and method using the features of the voice according to the present invention, by applying the voice dialing function to the existing mobile phone by software only without the addition of hardware, it is possible to perform speech recognition in a variety of noise environments such as highways This allows voice dialing not only in the handset but also in the handsfree kit. Therefore, it is possible to pursue the high quality of the terminal only by adding a memory for storing voice features, thereby increasing the competitiveness of the product. In addition, it is possible to recognize the voice relatively easily by using features of the voice such as the number of peaks of the voice belonging to the voice interval and / or the peak level of the voice.

Claims (2)

In the voice dialing apparatus using a feature of voice dialing by voice without pressing a button when dialing a mobile communication terminal, An input interface for extracting voice feature parameters from the frame-by-frame packet data output from the vocoder; A speech section detector for detecting a speech section by using the speech feature parameter output from the input interface; A feature extraction unit for extracting at least one of the number of peaks of the voice signal and the level of the peak in the detected voice section as a feature of the voice signal for voice recognition; A voice register that registers features of the voice recognition voice signal extracted from the feature extractor into a database so as to correspond to each telephone number; A voice recognition unit which selects the closest voice signal by comparing the voice signal registered in the database with the voice signal registered for the voice dialing using the feature of the voice recognition voice signal extracted from the feature extractor; And And a result output unit for outputting a voice signal and a phone number corresponding to the voice recognition result. A voice dialing method using a feature of voice dialing by voice when dialing a mobile communication terminal, Converting the packet stream signal output from the vocoder into an unpacking parameter stream signal; Detecting a section of the unpacking parameter stream signal; Detecting at least one of the number of peaks of the voice signal present in the detected signal and the level of the peak as a feature of the voice signal; A voice registration step of storing the feature of the voice signal in a memory so as to correspond to each telephone number; A voice recognition step of selecting a voice signal closest to the voice spoken from the registered voice signal; And And outputting a voice signal and a phone number corresponding to the voice recognition result.

Images