The Design And Experimental Study Of A Kind of Speech Instruction. Control System Prototype of Manned Spacecraft

Transcription

1 The Design And Eperimental Stud Of A Kind of Speech Instruction Control Sstem Prototpe of Manned Spacecraft Hao Zhai Xiaolin Yang Jianhua Yang LanZhou Institute of Phsics BOX 94, Lanzhou, P.R. China, Tel: , Fa: zhaihao8848@sina.com ABSTRACT The application of speech instruction control sstem in man machine interface of the manned spacecraft can enrich the intelligentization of the man machine interface and lighten the operation load of cosmonauts. Speech recognition integrated with speech snthesis constructs Speech Instruction Control Sstem (SICS). The design of a kind of Speech Instruction Control Sstem (SICS) prototpe is presented in this paper. The 32 bit embedded sstem is adopted as the hardware core, while the middle laer is embedded Linu. Outside the middle laer is the application software shell. The development and eperimental stud on the SICS prototpe has been carried out. Under the 00Mbps fast switch Ethernet and WINDOWS 98 OS environment, a simplified Chinese SICS bas been realized. In the sstem, a broadening endpoint Dnamic Time Warping (DTW) algorithm is applied which implements speaker-dependent rapid speech recognition and has the adaptive performance to process the tin diversification of the speech speed. The UDP protocal is adopted to implement the real time transmission of the recognition result code of speech commands. The SICS prototpe have been also realized in Linu Redhat 7.0. It is concluded from the eperiment that the SICS prototpe is feasible. There are several aspects of the problems pointed out in this paper for the SICS prototpe to be applied in man machine interface of the manned spacecraft. Introduction The application of speech recognition in man machine interface of the manned spacecraft can enrich the intelligentization of the man machine interface and lighten the operation load of spacemen. It is a profitable complementarit to the manual operation, especiall in the zero gravit environment. The Channel of the input speech is just like the third hand of the spaceman. Especiall, it can implement the remote operation during the walking out of the cabin for the spacemen (such as during maintenance, rendezvous etc.). In this paper, we bring forward a speech instruction control sstem (SICS) prototpe of manned spacecraft, and give the result of eperimental stud in a simplified SICS in which rapid speech recognition is applied to realize speech instruction control. 2 Design of A Speech Instruction Control Sstem Prototpe of Manned Spacecraft In man-machine interface of a spacecraft, speech recognition integrated with speech snthesis constructs a Speech Instruction Control Sstem (SICS). On one hand, the sstem can recognize the input speech command, perform the operation defined b the speech command and implement speech

2 control operations. On the other hand, the SICS can impart the operation acknowledgement cue according to the first recognition result to the spacemen with the snthesized speech. Furthermore, speech alarm of the parameter eceeding and event speech informing can be carried out if necessar. The primar design of a kind of Speech Instruction Control Sstem (SICS) prototpe is presented here. The embedded sstem is adopted as the hardware platform. In the platform, 32 bit MCU PowerPC is the hardware core while DSP and Codec Module is configured as the speech processing hardware, 8M bit flash and 8M bit RAM are epanded as storage media and high speed industrial ethernet interface instead of MIL-STD-553B is collocated to implement the sstem integration. The hardware block diagram is shown as Fig.. Above the hardware is RTOS embedded Linu. As a real time operating sstem, Linu is measured not onl b the correctness of the result but b the time in which the results are produced. Outside the middle laer is the application software shell. On the basis of RTOS, the demand for real-time of both speech recognition and speech snthesis is satisfied. Now das, information technologies have been rapidl developed. Wh we can not tr some new method, new device in our traditional space industr especiall in the manufacture of device level. So, new technolog of embedded sstem development and industrial devices are adopted in this prototpe promising the high reliabilit and low power consumption is insured. Thus new thinking for the future manned spacecraft development is provided. High Speed Industrial Ethernet Interface Flash and RAM Module The Embedded MCU Module DSP and Codec Module Earphone/Speaker Microphone/larngophone Fig. Hardware Block Diagram of SICS Prototpe Input Command Speech Recognition As Recognition Result,Output Snthesized Speech as Operation Cue Y Is It Sure N Recognize The Acknowledgement Command Recognize The Cancel Comma Eit Fig. 2 The Framework of SICS Software

3 The framework of the software is shown as Fig. 2. The speech recognition must be accurate, reliable, rapid and robust. Isolated word or discontinuous speech recognition can satisf the demand of the speech command control. Speaker-dependent and speaker-independent sstem can both satisf the requirement while multi-user templates storage technolog can get over the location of it. 6 Bit 32kpbs limited vocabular speech snthesis technolog is adopted to snthesize the speech for speech recognition feedback, speech alarm and speech informing. In addition, barge-in technolog must be used to make the SICS can process speech recognition during the snthesizing. The recognition result is transmitted to the OBDH via high speed serial interface. Meanwhile, the code of speech alarm and event is received from the interface. 3 The Development of Rapid Speech Recognition Sstem S(n) Filter Bank Pattern Training Templates Pattern Classifier (DTW algorithm) Decision Logic Recognition Result Fig. 3 Block Diagram of Pattern-Recognition approach The classic Pattern-Recognition approach is used in our development. The block diagram of the approach is shown as Fig. 3. In this paper, a 8-channel octave band filter bank is used in feature etraction phase of speech recognition which separates the signal frequenc bandwidth in a number of frequenc bands where the signal energ is measured. In the sstem, a relaed endpoint constraints Dnamic Time Warping (DTW) algorithm is applied which implements speaker-dependent rapid speech recognition and has the adaptive performance to process the tin diversification of the speech speed. The local continuit constraints with slope weighting is shown as Fig. 4. The dnamic programming recursion formula is following: Where: ( ) (, i ) D i D i, i : the minimum partial accumulated distortion along a path connecting (,) and (i,i) d ( i, i ): the short-time spectral distortions D( i 2, i ) + [ d( i, i ) + d( i, i )], 2 = min D( i, i ) + d( i, i ), D( i, i ) + [ d( i, i ) + d( i, i )] 2

4 /2 (i -2,i -) (i -,i ) /2 (i -,i -) (i,i ) /2 (i,i -) (i -2,i -) /2 The DTW algorithm is finding the best path through a T b T grid, beginning at (,) and ending at (T,T ), as follows:. Initialization where: m(k), the local slope weighting 2. Recursion For D A (,) = d(,) m() i T, i T such that i and i sta within the allowable grid, compute ς D A ( i in which: d( ( k), φ ( k) ) 3. Termination Fig. 4 The Local Continuit Constraints with slope weighting, i ) = min ( i ', i ' ) [ D ( i', i' ) + ς ( i', i' ), ( i, i ))] where: ( i', i' ), ( i, i ) A ( ) ς is defined b L s ( i ', i ' ), ( i, i ) ) = d ( φ ( T ' l ), φ ( T ' l )) l = 0 m ( T ' l ) φ is the short-time spectral distortions of φ ) and φ ) (k (k d ( X, Y ) = D A ( T, T ) M φ d, is the dissimilarit between X and Y, M φ is the normalizing factor where: ( X Y ) The following new set of boundar conditions is used to rela the endpoint constraints: φ ( ) + Q ma φ () + T Q ( T ) T ma φ T φ ( T ) T where: represents the maimum anticipated mismatch in the endpoint of the pattern

5 To reach a higher recognition rate, sentence sncopation and keword matching is applied in this sstem. The Modified K-means(MKM) algorithm is used in template training, and multi user templates storage technolog is used to get over the localization of speaker-dependence. That is, different speaker choice the different templates librar trained previousl. 4 Eperiment The development and eperimental stud on the SICS prototpe have been achieved. Under the 00Mbps fast switch Ethernet and WINDOWS 98 OS environment, a simplified Chinese SICS bas been realized. The configuration of eperimental environment of the SICS is shown as Fig. 5. The above Rapid Speech Recognition Sstem is applied in the SICS. The sample rate is.052khz. As a result, the responding time of the sstem is less than 0.5 second and the first recognition ratio of 4 speech control commands (including operation and operation object ) arrives at 95% under the laborator environment (less than 40dB). In addition, the mistake operation is avoided because of the using of operation acknowledgement of snthesized speech. The UDP protocal is adopted to implement the real time transmission of the recognition result code of speech commands. It is also implemented that the recognition result code drives each corresponding dela output in a 6 delas arra remotel to demonstrate the real time performance of speech control. Recentl, the SICS prototpe have been realized in Linu Redhat 7.0, drawing up the final destination and the eperiment result is the same. Server/Sstem Controller 00M Switch Ethernet Ethernet SICS Unit Drive Demonstartion Unit Rela Output Drive Card Microphone Sound Output Drive Demonstration Panel Fig. 5 The Configuration of Eperimental Environment of The SICS

6 5 Conclusion The speech instruction control sstem prototpe presented in this article can be used in space ship, space station or space lab, especiall the spaceman s walking out of the cabin. The main framework of the sstem is demonstrated viable. In speech control application in man machine interface of the manned spacecraft, vocabular is not the main factor, 200 vocabular is much enough. The main factor are speed (real time performance), recognition rate and robust. Finall, there are several aspects of the problems which must be resolved according to the SICS prototpe: more high recognition rate robust sstem further miniaturization the application of larngophone References [] Lawrence Rabiner, Biing-Hwang Juang, Fundamentals of Speech Recognition,993. [2] Richard L. Klevans, Robert D.Rodman, Voice Recognition. [3] Brain Wanstall, DVI in The Militra Cockpit A Third Hand For The Combat Pilot. [4] Jianhua Yang, Hao Zhai, A Kind of Chinese Speech Snthesis Sstem with Large Vocabular Which Can Be Used in Man-Machine Interface of Spacecraft,IAF-99-U [5] Hao Zhai, Jianhua Yang, A Stud on the Application of Speech Snthesis Technolog with Limited Vocabular in Man-Machine Interface of Manned Spacecraft, 8 th ISCOPS,999.