Baseband modem and method for speech recognition and mobile communication terminal using the same

<u style="single">BACKGROUND OF THE INVENTION</u>

<u style="single">Field of the Invention</u>

The present invention relates to a baseband modem and method for speech recognition, and more particularly, to a baseband modem and method for speech recognition and a mobile communication terminal using the baseband modem and method. Although the present invention is suitable for a wide scope of applications, it is particularly suitable for securing a higher rate of speech recognition.

<u style="single">Description of the Related Art</u>

Generally, a conventional baseband modem includes an audio codec. Conventional speech recognition technology, as applied to a mobile communication terminal, generally utilizes the same sampling rate for both vocoding of voice communication and voice recognition. The same sampling rate is utilized because there are few baseband modems capable of supporting an input of a 16kHz microphone and most baseband modems have difficulty obtaining PCM (pulse code modulation) data.

FIG. 1 is a block diagram illustrating a conventional baseband modem. FIG. 2 is a flowchart illustrating a conventional speech recognition method utilizing the baseband modem illustrated in FIG. 1.

Referring to FIG. 1, a conventional baseband modem includes an audio codec 13, a vocoder 15 and a processor 17. Once a voice signal is received from a microphone, the audio codec 13 performs modulation on the voice signal at a prescribed sampling rate. For example, PCM (pulse code modulation) is performed on the voice signal at a sampling rate of 8kHz.

The vocoder 15 performs vocoding on an output of the audio codec 13. For instance, QCELP (Qualcomm Code Excited Linear Prediction) or EVRC (Enhanced Variable Rate Coding) is performed.

The processor 17 performs speech recognition on an output of the vocoder 15. Specifically, the processor 17 decodes vocoded data and then extracts a feature vector from the decoded data. The processor 17 performs speech recognition by applying the extracted feature vector to a speech recognition algorithm that was previously prepared. Preferably, the processor 17 includes an MPU (micro processing unit) or DSP (digital signaling processor). On the other hand, if the voice signal is for voice communication, the processor 17 performs channel encoding, using either a convolution code or turbo code, on the output of the vocoder 15.

A conventional speech recognition method according to the above-explained configuration is explained with reference to FIG. 2.

Once a voice signal is received from a microphone, the conventional baseband modem performs modulation on the voice signal at a prescribed sampling rate (S12). For example, PCM (pulse code modulation) is carried out on the inputted voice signal at a sampling rate of 8kHz.

Vocoding of the modulated voice signal is then performed (S14). For example, QCELP (Qualcomm Code Excited Linear Prediction) or EVRC (Enhanced Variable Rate Coding) is utilized for vocoding.

Speech recognition is performed on the vocoded signal in an MPU (micro processing unit) or DSP (digital signaling processor). For speech recognition, vocoded data is decoded (S16) and a feature vector is extracted from the decoded data (S18). The extracted feature vector is then applied to a speech recognition algorithm (S20).

In the conventional method, the sampling rate for modulation is set to 8kHz. This is because a speech level of a quality that is recognizable can be provided using a voice component below 4kHz.

However, when performing speech recognition in a mobile communication terminal according to the conventional method, data processed according to sampling for voice communication is used. Therefore, the conventional method is unable to guarantee a satisfactory speech recognition rate. Furthermore, in the conventional method, unnecessary vocoding and decoding are performed as illustrated in FIG. 2.

Optionally, a digital signal processing chip or a speech recognition chip for speech recognition may be included in the mobile communication terminal. However, this increases the cost of a terminal.

In some conventional baseband modems, a method such as DTW (dynamic time warping) has been used for speech recognition. Since the data is processed according to sampling for voice communication, this method fails to guarantee a satisfactory speech recognition rate. In the conventional speech recognition method, either the sampling rate of the audio codec provided in the baseband modem is increased or extracting of the feature vector is not implemented with hardware.

There is another conventional method for speech recognition. In this method, a separate audio codec having a sampling rate suitable for speech recognition is installed outside the baseband modem. However, the corresponding hardware implementation is very complicated.

Conventional mobile communication terminals that perform speech recognition are unable to adjust the sampling rate of the baseband modem by separating voice communication from speech recognition. Furthermore, conventional baseband modems have difficulty obtaining the PCM (pulse code modulation) data.

<patcit id="pcit0001" dnum="US6411926B1"><text>US 6,411,926 B1</text></patcit> discloses a voice recognition system for use, inter alia, in a wireless telephone and including a digital signal processor configured to extract parameters from digitized input speech samples and provide the extracted parameters to a microprocessor. The microprocessor compares the parameters with stored template vocabulary contents to determine the stored word that most closely resembles the utterance. Having recognized the utterance, the microprocessor performs a command that is associated with the stored word, such as, e.g., initiating a telephone call. In one embodiment, the parameters extracted by the digital signal processor are power values of the digitized speech samples in a plurality of frequency bands. It is disclosed in this document that the digital signal processor may be a conventional vocoder. In this case, the extracted parameters may be diagnostic data generated by the vocoder. In any case, the digital signal processor also performs speech encoding for voice communication.

<patcit id="pcit0002" dnum="US5166971A"><text>US 5,166,971</text></patcit> discloses a telephone switching system capable of performing speaker recognition based on digitized speech samples delivered from telephone terminal equipment connected to the switching system via ISDN connections. The terminal equipment includes an A/D converter operable at different sampling frequencies, a standard sampling frequency for voice communication and a higher sampling rate for speaker recognition. Speech specimens sampled at the higher sampling rate are intermediately stored in the terminal equipment, are read out with the standard sampling frequency and are transmitted to the switching system in which the actual speaker recognition is carried out.

<patcit id="pcit0003" dnum="US6321195B1"><text>US 6,321,195 B1</text></patcit> discloses a mobile terminal capable of automated dialling. The telephone includes an audio codec for PCM modulation of speech input into the telephone, and a vocoder for vocoding the PCM-encoded speech samples. The output data of the vocoder include line spectrum pair (LSP) parameters that are used by a controller of the telephone for speech recognition. It is stated in this document that due to the vocoder providing the LSP coefficients, no special process for extracting features for speech recognition is required.

It is an object of the present invention to provide a baseband modem and a method of performing speech recognition and speech communication in such baseband modem that allow the use of an optimized sampling rate for speech recognition to guarantee a satisfactory speech recognition rate without performing unnecessary vocoding and decoding.

To address the above objects, the present invention provides a baseband modem according to claim 1 and further provides a method of performing speech recognition and speech communication in a baseband modem according to claim 18. Preferred embodiments are given in the dependent sub-claims.

The invention is directed to a baseband modem and method for speech recognition and a mobile communication terminal using the baseband modem and method. By using a variable sampling rate, a rate optimized for speech recognition is utilized in order to secure a higher rate of speech recognition.

The baseband modem includes an audio codec adapted to modulate a voice signal using one of a first sampling rate and a second sampling rate, means for speech recognition and means for speech encoding. The audio codec encodes the voice signal using the first sampling rate and speech recognition means performs speech recognition of the encoded voice signal if the voice signal is a voice command and the audio codec encodes the voice signal using the second sampling rate and the speech encoding means performs vocoding of the encoded voice signal if the voice signal is voice communication.

Preferably, the speech recognition means includes a feature vector extraction block adapted to extract one or more feature vectors from the encoded voice signal and a speech recognition block adapted to perform speech recognition using an extracted feature vector. It is contemplated that the speech recognition block includes a buffer adapted to store the feature vectors extracted from the encoded voice signal.

It is contemplated that a buffer is provided to store the encoded voice signal, for example, a ping-pong buffer. Preferably, the feature vector extraction block extracts the feature vectors from data stored in the buffer.

Preferably, the feature vector extraction block is implemented in hardware. Alternately, the feature vector extraction block may be implemented in software.

The baseband modem includes a controller to determine whether the voice signal is a voice command or voice communication. The controller may power on registers of the feature vector extraction block and speech recognition block if the voice signal is a voice command and powers off registers of the feature vector extraction block and speech recognition block if the voice signal is voice communication. The controller determines the sampling rate used by the audio codec.

Preferably, the speech encoding means includes a vocoder adapted to vocode the encoded voice signal. It is contemplated that the second sampling rate is optimized for voice communication, for example, 8 kHz.

Preferably, the first sampling rate is optimized for speech recognition. It is contemplated that the first sampling rate is in a range of approximately 12 kHz to approximately 32 kHz, for example, 16 kHz.

Preferably, the audio codec perform pulse code modulation on the voice signal. Preferably, the baseband modem is implemented in a mobile communication terminal.

The method of performing speech recognition and speech communication in a baseband modem includes determining whether a voice signal is a voice command or voice communication and modulating the voice signal using a first sampling rate and performing speech recognition of the modulated voice signal if the voice signal is determined to be a voice command and modulating the voice signal using a second sampling rate and performing vocoding of the modulated voice signal if the voice signal is determined to be voice communication.

Preferably, speech recognition is performed by extracting one or more feature vectors from the modulated voice signal and performing speech recognition using an extracted feature vector. It is contemplated that the extracted the feature vectors may be stored in a buffer.

It is contemplated that the modulated voice signal may be stored in a buffer. Preferably, the feature vectors are extracted from data stored in the buffer.

Preferably, feature vector extraction is implemented in hardware. Alternately, feature vector extraction may be implemented in software.

Determining whether the voice signal is a voice command or voice communication is performed according to a user selection. It is contemplated that activation of a feature vector extraction block and a speech recognition block may be controlled such that the feature vector extraction block and speech recognition block are activated if the voice signal is a voice command and the feature vector extraction block and speech recognition block are deactivated if the voice signal is voice communication. Preferably, registers of the feature vector extraction block and speech recognition block are powered on if the voice signal is a voice command and are powered off if the voice signal is voice communication.

It is contemplated that the voice signal is modulated at a first sampling rate optimized for speech recognition. It is contemplated that the first sampling rate is in a range of approximately 12 kHz to approximately 32 kHz, for example, 16 kHz.

It is contemplated that the voice signal is modulated at a second sampling rate optimized for voice communication. Preferably, an 8 kHz rate is used.

Preferably, pulse code modulation is performed on the voice signal. Preferably, the baseband modem is implemented in a mobile communication terminal.

Additional features and advantages of the invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the invention. It is to be understood that both the foregoing general description and the following detailed description of the present invention are exemplary and explanatory and are intended to provide further explanation of the invention as claimed.

These and other embodiments will also become readily apparent to those skilled in the art from the following detailed description of the embodiments having reference to the attached figures, the invention not being limited to any particular embodiments disclosed.

<u style="single">BRIEF DESCRIPTION OF THE DRAWINGS</u>

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention. Features, elements, and

FIG. 1 is a block diagram illustrating a conventional baseband modem.

FIG. 2 is a flowchart of a conventional speech recognition method utilizing the baseband modem illustrated in FIG. 1.

FIG. 3 is a block diagram of a baseband modem according to one embodiment of the present invention.

FIG. 4 is a flowchart of speech recognition method according to one embodiment of the present invention.

<u style="single">DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS</u>

The present invention relates to a baseband modem and method for speech recognition and a mobile communication terminal using the baseband modem and method. Although the present invention is illustrated with respect to a mobile communication device, it is contemplated that the present invention may be utilized anytime it is desired to perform voice recognition and voice communication using optimized sampling rates in order to secure a higher rate of speech recognition.

Reference will now be made in detail to the preferred embodiments of the present invention, examples of which are illustrated in the accompanying drawings. Wherever possible, the same reference numbers will be used throughout the drawings to refer to the same or like parts.

A baseband modem for voice recognition and mobile communication terminal using the baseband modem according to a preferred embodiment of the present invention is explained with reference to FIG. 3. FIG. 3 is a block diagram illustrating a baseband modem according to one embodiment of the present invention, in which the baseband modem is preferably provided in a mobile communication terminal. Referring to FIG. 3, a baseband modem includes an audio codec 22, a controller 27, a vocoder 28, a feature vector extraction block 24, a plurality of buffers 23 and 25 and a speech recognition block 26.

When a voice signal is received from a microphone, the audio codec 22 performs modulation on the inputted voice signal at a selected sampling rate. The microphone transforms a user voice into an electrical signal. Specifically, the audio codec 22 performs PCM (pulse code modulation) on the voice signal at a selected sampling rate.

The audio codec 22 changes the sampling rate to perform the PCM according to whether the voice signal corresponds to a signal for speech recognition or a signal for voice communication. Specifically, the audio codec 22 applies a sampling rate of approximately 8kHz to the PCM performed on the voice signal for voice communication. On the other hand, the audio codec 22 applies a sampling rate of 12~32kHz to the PCM performed on the voice signal for speech recognition.

Preferably, the audio codec 22 applies a sampling rate of 16kHz to the PCM performed on the signal for speech recognition. This is because it is known that a sampling rate of 16kHz enhances a speech recognition rate.

A user selects an application to identify whether the voice signal corresponds to a signal for speech recognition or a signal for voice communication. Specifically, if the user selects the application for voice communication, a signal received by the audio codec 22 thereafter corresponds to a voice signal for voice communication. If the user selects the application for speech recognition, a signal received by the audio codec 22 thereafter corresponds to a voice signal for speech recognition.

In the present invention, by determining what type of the application the user selects, the controller 27 activates either a signal transfer path for voice communication or a signal transfer path for speech recognition. Specifically, the controller 27 activates or deactivates elements 23, 24 and 25 of the signal transfer path for speech recognition.

If the user selects the application for speech recognition, the controller 27 activates elements 23, 24 and 25 of the signal transfer path for speech recognition. If the user does not select the application for speech recognition, the controller 27 deactivates elements 23, 24 and 25 of the signal transfer path for speech recognition to cause the output of the audio codec 22 to be transferred to the vocoder 28.

Furthermore, the controller 27 controls the sampling rate of the audio codec 22. Specifically, the controller 27 can determine whether the signal received by the audio codec 22 is for voice communication or for speech recognition according to what type of application the user selects. The controller 27 controls the audio codec 22 to perform the PCM using the sampling rate suitable for each type of application.

An example of a control operation of the controller 27 is explained as follows. Once a user selects an application for speech recognition in order to perform, for example, auto-dialing, menu selection or name paging, the controller 27 powers on particular registers of the baseband modem used for a speech recognition mode. The controller 27 sets the sampling rate of the audio codec 22 to a speech recognition sampling rate, for example, 16kHz. The coder 27 then powers on the portion of the baseband modem utilized for speech recognition mode, specifically buffer 23, feature vector extraction block 24 and feature vector buffer 25.

In brief, the controller 27 varies the sampling rate used by the audio codec 22 and determines a path for transfer of the output of the audio codec 22 according to the application selected by the user.

In the signal transfer path for speech recognition, an output of the buffer 23 is provided to an input of the feature extraction block 24. The buffer 23 stores a voice signal (PCM data) for speech recognition. Preferably, the buffer 23 is a ping-pong buffer.

Specifically, the ping-pong buffer uses a double buffering structure. In a double buffering structure divided into two storage areas, one of the two storage areas stores data while the other storage area outputs the data stored in the former storage area. Preferably, the present invention uses the double buffering structure or a structure including at least three divided storage areas configuring a ring shape. Furthermore, the buffer 23 includes a 20~40ms buffer.

The feature vector extraction block 24 receives the PCM data from the buffer 23 and extracts feature vectors from the received PCM data. The feature vector extraction block 24 adopts MFCC (mel-frequency cepstral coefficients), PLP (perceptual linear prediction), LPC (linear predictive coding) or LPCC (linear predictive cepstral coefficients). A feature vector buffer 25 stores the feature vectors extracted by the feature vector extraction block 24. In the present invention, the feature vectors are repeatedly extracted by a short time unit of 20~40ms and the extracted feature vectors are stored in the feature vector buffer 25 in the form of an array.

Generally, when extracting feature vectors, filter- bank filtering, FFT (fast Fourier transform), DCT (discrete cosine transform) and IFFT (inverse fast Fourier transform) should be conducted. Therefore, a large volume of operations is required for extracting the feature vectors and the feature vector extracting process has strong repeatability.

Preferably, the present invention implements the feature vector extraction block 24 in hardware. However, the feature vector extraction may be implemented in software.

The speech recognition block 26 performs speech recognition using the feature vectors stored in the feature vector buffer 25. Preferably, the speech recognition block 26 includes an MPU (micro-processing unit) or DSP (digital signaling processor) provided with a speech recognition algorithm.

The variability of a speech recognition algorithm is very high. A difference of fixed point implementation may exist according to a training file and parameters. Parts corresponding to Viterbi decoding, language modeling or grammar for the enhancement of the algorithm are used. Therefore, the parts for fixed point implementation or algorithm enhancement in the speech recognition algorithm are implemented via the aforementioned MPU or DSP.

Furthermore, noise cancellation may be performed in the present invention for speech recognition via the MPU or DSP. Preferably, the noise cancellation is executed via the MPU or DSP.

The vocoder 28 performs vocoding on the output (PCM data using the sampling rate of 8kHz) of the audio codec 22 for voice communication. Specifically, if a voice signal for voice communication is received, the vocoder 28 performs the vocoding using QCELP (Qualcomm code excited linear prediction), EVRC (enhanced variable rate coding), VSELP (vector sum excited linear prediction) or RPE-LTP (residual pulse excitation/long term prediction). Channel coding is performed on an output of the vocoder 28 using convolution code or turbo code. Radio modulation is performed after completion of the channel coding.

FIG. 4 illustrates a method for performing speech recognition according to the present invention. The method includes receiving a voice signal (S100), determining whether the voice signal is a voice command or voice communication (S102) and either modulating the voice signal using a rate optimized for speech recognition (S104) and storing the modulated voice signal (S106), extracting a feature vector from the modulated voice signal (S108), storing the extracted feature vector (S110) and performing speech recognition using the extracted feature vector (S112) or modulating the voice signal using a rate optimized for voice communication (S114) and vocoding the modulated voice signal (S116).

Preferably, extracting a feature vector from the modulated voice signal (S108) is implemented in hardware. Alternately, extracting a feature vector from the modulated voice signal (S108) may be implemented in software.

Preferably, the determination of whether the voice signal is a voice command or voice communication (S102) is performed according to a user selection of a type of application. Preferably, pulse code modulation of the voice signal is performed.

Preferably the selection of one of the two paths (S104 - S112 and S114 - S116) is performed by controlling particular registers related to the feature vector extraction and speech recognition. Specifically, the registers related to the feature vector extraction and speech recognition are activated by applying power if the voice signal is determined to be a voice command (S102) and are deactivated by removing power if the voice signal is determined to be voice communication.

If the voice signal is determined to be a voice command (S102), a rate of approximately 12 kHz to approximately 32 kHz is used for modulating the voice signal, preferably 16 kHz. If the voice signal is determined to be voice communication (S102), preferably a rate of 8 kHz is used for modulating the voice signal.

Preferably, the baseband modem is included in a mobile communication terminal as an internal element when the mobile communication terminal is manufactured. Alternatively, the baseband modem may be implemented as an independent module to be assembled as part of a mobile communication terminal layer. Therefore, it can be understood that the scope of the present invention covers both of the aforementioned alternatives.

The present invention provides several effects or advantages. First, since a sampling rate suitable for speech recognition is utilized when modulation is performed by the audio codec, the speech recognition rate can be enhanced. Second, by implementing the feature vector extraction with hardware, the present invention can reduce the volume of operations of the processing unit for speech recognition and reduce the power consumption. Third, by implementing the fixed point implementation or the algorithm enhancement with the MPU or DSP in the speech recognition algorithm, the present invention facilitates expansion according to future necessity.