Voice recognition system and method thereof

This application claims the benefit of U.S. Provisional Application Ser. No. 60/842,693, the entire disclosure of which is hereby incorporated by reference.

1. Field of the Invention

The invention relates to voice recognition, and in particular to a communication system providing voice recognition and a method thereof.

2. Description of the Related Art

A voice recognition controller is a device with the ability to recognize keywords provided by a user and generate corresponding outputs, for example, commands or text-strings, for use by an application device, such as DVD/MP3 players, mobile phones, car radio, and car controllers.

The voice recognition controller may be a Bluetooth speakerphone for use with a mobile telephone provided with Bluetooth functionality in a communication system, in which the voice controller speakerphone and mobile telephone are paired. A voice recognition engine on the voice controller is implemented to recognize a name with reference to a user-defined name list and output a corresponding telephone number. A dialing function on the mobile telephone will then dial the number, and the user is able carry on a conversation through the mobile telephone via the speakerphone.

There is a need for a communication system employing a voice recognition controller with improved voice recognition rate.

A detailed description is given in the following embodiments with reference to the accompanying drawings.

According to the invention, a communication system comprises a setup controller, a voice recognition controller, and an application controller. The setup controller receives a voice keyword table including a voice recognizable keyword and a corresponding application instruction. The voice recognition controller, coupled to the setup controller, receives the voice keyword table from the setup controller, receives a first voice input, and matches a first voice input to the voice recognizable keyword to determine the corresponding application instruction. The application controller, coupled to the voice recognition controller, receives and performs the corresponding application instruction. The setup controller, the voice recognition controller, and the application controller communicate through wireless communication.

A method of voice recognition in a communication system is also disclosed, comprising a setup controller receiving a voice keyword table including a voice recognizable keyword and a corresponding application instruction, a voice recognition controller receiving the voice keyword table and matching a first voice input thereto to determine the corresponding application instruction, and an application controller receiving and performing the corresponding application instruction.

The invention can be more fully understood by reading the subsequent detailed description and examples with references made to the accompanying drawings, wherein:

FIG. 1 is a block diagram of an exemplary communication system of the invention.

FIGS. 2 a, b and c illustrate exemplary voice keyword tables incorporated in the communication system of FIG. 1 .

FIG. 3 illustrates an exemplary method of text-to-speech (TTS) conversion and voice recognition models generation of the invention.

FIG. 4 is a block diagram of the voice recognition device in FIG. 1 .

FIG. 5 is a flowchart of a voice recognition method of the invention.

The following description is of the best-contemplated mode of carrying out the invention. This description is made for the purpose of illustrating the general principles of the invention and should not be taken in a limiting sense. The scope of the invention is best determined by reference to the appended claims.

FIG. 1 is a block diagram of an exemplary communication system of the invention, comprising setup controller 10 , voice recognition controller 12 , and application controller 14 . Setup controller 10 is coupled to voice recognition controller 12 , and to application controller 14 .

Setup controller 10 may be a personal computer or personal digital assistant (PDA). Setup controller 10 comprises voice keyword table VKT 100 , comprising voice recognizable keywords and corresponding application instructions. Voice keyword table VKT 100 may contain at least 1 set of voice keyword Vi and application instruction Ai pair. Voice recognition is performed by searching an input voice data from all voice keywords Vi. Voice recognition is classified by three general classes, namely speaker independent (SI), speaker dependent (SD) and speaker adapted (SA). In the SI system, a voice recognition engine identifies utterances according to universal acoustic models generated from samples obtained from a large training population. An example of commercial software application using SI models is Microsoft Speech Recognizer. As no individual training by the user is required, such systems are convenient. However, these systems generally have low recognition performance, especially when used by speakers with heavy accents or whose speech patterns otherwise diverge from the training population. On the other hand, SD systems require users to provide samples for every keyword, which can become burdensome and memory intensive for large lists of keywords. Voice keyword table VKT 100 may be edited by input such as a keyboard or a keypad, or imported from an established database such as a phonebook in an application such as Outlook, Skype, MSN, or Yahoo Messenger.

FIG. 2 a illustrates the exemplary voice keyword table incorporated in the communication system in FIG. 1 , comprising label field 20 a , adaptive voice keyword field 22 a , default voice keyword field 24 a , and application field 26 a . Label field 20 a contains name tags L 1 , L 2 , . . . , L i , . . . , and L n entered by the user. Each name tag Li may be converted to default voice keyword V Di through a speaker independent (SI) text-to-speech (TTS) algorithm in setup controller 10 . Default voice keywords are voice recognizable keywords to which an input voice data stream is compared. Default keyword field 24 need not be viewable. Setup controller 10 comprises a microphone for voice input. The user may further enter corresponding adaptive voice keywords V A1 to V An in adaptive voice keyword field 22 as alternative voice recognizable keywords. Adaptive voice keyword K ai may be adaptive to individual pronunciation parameters. Application field 26 provides application instruction A P1 for execution upon recognition of default voice keyword V A1 or adaptive voice keyword V A1 in setup controller 10 .

Referring to FIG. 1 , voice recognition controller 12 receives voice keyword table VKT 100 from setup controller 10 , and receives first voice input V 1 to perform first layer voice recognition (VR), i.e., seeking first voice input V 1 from the voice recognizable keywords in voice keyword table VKT 120 , thereby determining the corresponding application instruction. After locating first voice input V 1 in voice keyword table VKT 120 , voice recognition controller 12 transmits the corresponding application instruction to application controller 14 for execution. Voice recognition controller 12 may be a low-profile portable device such as a headset, speakerphone, or headset with VR functionality. Application controller 14 may be a cellular phone, DVD player, MP3 player, car radio, car controller, or any electronic consumer appliance.

Voice recognition controller 12 transmits first voice input V 1 to application controller 14 if no match is found for the first voice input in the voice keyword table. Application controller 14 further performs second layer voice recognition, i.e., determining a corresponding application instruction based on first voice input V 1 , and transmits first voice input V 1 and the corresponding application instruction to voice recognition controller 12 . The second layer voice recognition may be implemented by a local voice keyword table comprising other voice keywords and the corresponding applications in application controller 14 , such that first voice input V 1 is matched to one of the other voice keywords, thereby finding and performing the corresponding application A 1 . Application controller 14 then executes the corresponding application A 1 . The second layer voice reorganization recognizes more voice keywords than the first in voice reorganization controller 12 . Application controller 14 may further feed first voice input V 1 the corresponding application A 1 back to voice recognition controller 12 so that voice recognition controller 12 can recognize first voice input V 1 , thereby improving voice reorganization rate.

Setup controller 10 , voice recognition controller 12 , and application controller 14 communicate through wired or wireless communication. The wireless communication may be, for example, Bluetooth, Wireless LAN, WiFi, or equivalent.

Setup controller 10 , comprising self diagnostic function, receives second voice input V 2 for comparison with a voice recognizable keyword in voice keyword table VKT 120 , and validates the voice recognizable keyword if second voice input V 2 matches the voice recognizable keyword, allowing the user to change the voice recognizable keyword in setup controller 10 .

Voice keyword table VKT may be a phonebook, voice recognizable keyword is a pronunciation of a name entry, and the corresponding application instruction dialing a predetermined number. FIG. 2 b shows an exemplary voice keyword table VKT in FIG. 1 . Label fields 20 b contain user-defined name entries, for example, “Brian Smith” and “ ” Default voice keyword field 24 b comprises machine-generated phonetic data including phoneme string converted from the name entries through word-to-phoneme algorithm and the corresponding speaker independent models for speech recognition and text-to-speech generated voice. Adaptive voice keyword field 22 b includes user-defined phonetic data, possibly different from the phonetic data in default voice keyword field 24 b , it provides capability for users to further personalize data defined in 24 b to enhance the voice recognition performance and voice prompt quality of the system. The corresponding application instructions are text-strings, such as dialing the telephone numbers in the application filed 26 b.

Voice keyword table VKT may be a voice command table, the voice recognizable keyword pronunciation of a voice command entry, and the corresponding application instruction of a built-in command or a user-defined command. FIG. 2 c shows another exemplary voice keyword table VKT in FIG. 1 . Label field 20 c stores command labels such as “HOME”, “VOICEMAIL”, “SPEEDIAL[1-9]”, “MEMORY[1-9]”, or “SETUP MENU”, default voice keyword field 24 c contains the corresponding default phonetic data, adaptive voice keyword field 22 c are the adaptive phonetic data. Both the default and adaptive phonetic data are voice recognizable keywords correspond to specific commands. For example, the default phonetic or adaptive phonetic data corresponding to “HOME” maps to the application instruction of dialing a home number, “VOICEMAIL” maps to dialing a voicemail number, “SPEEDIAL[1-9]” maps to dialing a number according to the speed dial digit 1 to 9, “MEMORY[1-9]” maps to dialing a number in memory 1 to 9, and “SETUP MENU” maps to entering the setup menu.

FIG. 3 illustrates an exemplary method of text-to-speech (TTS) conversion and voice recognition in the invention, incorporating the voice keyword table VKT in FIG. 2 , and the setup controller in FIG. 1 . Method 3 improves the voice recognition performance by sharing the same phonemes between TTS voice prompt and voice recognition recognition models. Setup controller 10 will generate proper pronunciation phoneme and user can edit the pronunciation phoneme under the guide of TTS prompt and according his/her habitual pronunciation, once the pronunciation phoneme is confirmed, system will further generate VR model for recognition. Thus the voice recognition performance can be enhanced based on user's habit. The method in FIG. 3 uses common phoneme strings in TTS and voice recognition, so that the setup controller can produce voice keyword table VKT that is adaptive to the user's pronunciation preference thus enhance the recognition performance. The steps above the dotted line provide text-to-speech conversion, and those below for voice recognition model generation.

In step S 310 , setup controller 10 receives text stream “Home” from S 300 for example, the label field 20 a in FIG. 2 a , and generate pronunciation phonemes for “Home” S 312 out of all phonemes S 302 of the language. The word-to-phoneme algorithm S 310 which may search a pronunciation dictionary or apply pronunciation generation rules for out-of-dictionary text streams.

In step 320 , Text-to-Speech module generates speech waveform S 322 for “Home” according to phoneme string S 312 .

In step S 330 , setup controller 10 receives the phonemes S 312 to generate a voice data based on the full set built-in speaker independent models S 332 . For example, system controller 126 fetches “h” “o” “m” models in S 332 according to S 312 in to generate the corresponding voice recognition models. S 334 for “Home”.

In step S 340 , setup controller 10 employs the speaker dependent speech data S 342 and the speaker dependent models S 334 to generate speaker dependent voice recognition models S 344 . Since speaker dependent models accommodates habitual pronunciation of the user, therefore the voice recognition rate is improved.

FIG. 4 is a block diagram of the voice recognition device in FIG. 1 , comprising first wireless transceiver 120 , second wireless transceiver 121 , voice input device 122 , voice output device 123 , analog to digital converter 124 , memory 125 , system controller 126 , and speaker 127 .

First wireless transceiver 120 downloads VKT 100 from setup controller 10 to store in memory 125 . Second wireless transceiver 121 transmits the corresponding application instruction Ai to and receives requested voice recognition keyword Vi and the corresponding application Ai from application controller 14 , added into VKT 120 subsequently. Voice input device 122 may be a microphone, receiving first voice input V 1 . Voice output device 123 may be a speaker playing voice data. Analog to digital converter (ADC) 124 converts first voice input V 1 from analog to digital. Memory 125 stores voice keyword table VKT 120 , built-in speech models and parameters, and first voice input V 1 . System controller (voice recognition module) 126 matches first voice input V 1 to voice keywords V i in table VKT to determine the corresponding application instruction Ai.

FIG. 5 is a flowchart of a voice recognition method in the invention, incorporating the voice recognition controller in FIG. 4 .

In step S 500 , noise in first voice input V 1 is suppressed in time domain to provide good signal quality.

In step S 502 , first voice input V 1 is converted to frequency domain to further reduce noise and extract speaker specific models therein.

In step S 504 , system controller 126 uses first voice input V 1 , speaker independent and speaker dependent speech models to search the voice recognizable keywords in table VKT.

In step S 506 , system controller 126 performs post-processes on first voice input V 1 , such as, for example, execution of corresponding application instructions, first layer rejection, or generation of the success rate of voice recognition. If system controller 126 determines a first voice input V 1 match with a voice recognition keyword in table VKT, the corresponding application instruction is transmitted to application controller 14 for execution. If not, system controller 126 transmits first voice input V 1 to application controller 14 for second layer voice recognition. System controller 126 may establish a score system based on success rate of the first layer voice recognition.

FIG. 4 While the invention has been described by way of example and in terms of preferred embodiment, it is to be understood that the invention is not limited thereto. To the contrary, it is intended to cover various modifications and similar arrangements (as would be apparent to those skilled in the art). Therefore, the scope of the appended claims should be accorded the broadest interpretation so as to encompass all such modifications and similar arrangements.