Abstract: Introduced one complete based on PC machine and DSP (TMS320C50) general voice signal processing system. This system may realize the different voice signal analysis and the processing algorithm through the programming design, can aim at the different application and the new processing method, increases the DSP program module unceasingly, improves and the expansion system’s function by this. Moreover, this system has also provided the high accuracy pronunciation data acquisition and the pronunciation data playbacking function, and the overall system has the good versatility and the high operating efficiency.
Key word: Pronunciation processing digital signal processing TMS320C50 chip
The language is the humanity communicates the information mutually important labor. Along with the modern science technology’s development, specially the voice communication and each kind of pronunciation product’s widespread popularization, voice signal’s digitized processing is playing the huge role in more and more domains. At present, each kind already entered the market take the speech signal digital processing as the characteristic commodity, the commercialized voice signal processor has also been published, like KAY Corporation’s CSL TM (Computerized Speech Lab).
Not only a complete voice signal processing system must have voice signal gathering and the playbacking function, moreover must be able more importantly to complete the complex voice signal analysis and the processing algorithm. Usually these algorithm operand is big, and must satisfy real-time or the real-time fast highly effective processing request, must therefore use the high speed DSP chip. Moreover, during the request system satisfies good versatile, and presents the new processing method unceasingly in view of the different application, but must cause the system to be advantageous for the function the improvement and the expansion.
Therefore, we take PC machine as the main engines, as the signal processing core has designed this system take TMS320C50, its hardware disposition diagram see Figure 1. And, TMS320C50 is American Texas Instrument Corporation’s 16 fixed-point DSP product, it including improvement Harvard (Harvard) the structure, high performance CPU, the internal memory, highly effective releases in the piece periphery connection as well as a set arranges the set of instructions, the computation speed may reach 40Mips, and performance price compared to good.
1 system structure
The system hardware sets at as shown in Figure 1.
1.1 PC main engines
Considered system’s versatility and the easy possibility, we to take CP machine system’s main engine, its concrete disposition by different application determination. It mainly provides system’s man-machine conversation contact surface, controls completes system each function.
1.2 PREPROC parts
The overall system uses the entire modular structure, the disposition is nimble, the debugging maintenance is convenient, supports each kind of pronunciation digital processing algorithm which the software realizes. The overall system and four module parts is composed of PC machine. These four module part respectively be input signal enlargement part (PREPROC), power amplification part (POSPROC), high speed figure signal processing part (DSP) and 16 high accuracy sampling playbacking part (AD/DA). And, the DSP part completes the concrete signal processing task, and controls the AD/DA part to complete the voice signal gathering and playbacking; The PREPROC part completes to inputs the voice signal the enlargement and against aliasing filter; The POSPROC part completes the smooth filter and the power amplification.
This part we the input signal carries on the enlargement and against aliasing filter processing to the microphone input signal or the line, its output supplies the AD/DA part to carry on the sampling. Its design target is: Microphone input range 1mV~31.6mV, line input scope 100mV~3.16V, the input impedance 10kΩ, the part increases adjustable, may enlarge in the nominal range the signal to the AD input full scale division ±10V. In which against aliasing filter take the independent plug-in unit, the detailed design showed in behind.
1.3 POSPROC parts
This part the simulated signal which sends out to D/A carries on restores the filter and the power amplification, its output supplies field sound. May output the maximum work rate is 1W, uses for impels 8Ω speaker. And restores the filter to take the independent plug-in unit, the detailed design showed under.
1.4 filter plug-in units
This system provides 10kHz and the 20kHz two kind of filter plug-in units supplies the choice. For the adaptation different application need, the system may dispose the different cut-off frequency the filter plug-in unit or the external connection cut-off frequency adjustable filter.
The system provides against aliasing filter and restore the filter, separately two eight step Chebyshev low pass filter series constitution which constitutes by two piece of MAX274 and the external connection resistances. May know by the component handbook, MAX274 is composed of four same filter units, each filter unit is one second-order filters the filter. For convenience, numbers in turn four filter units A, B, C, D. May look up the pass band fluctuation by “Active Filter Precise Design” is normalized parameter B which and C the 0.2dB eight step Chebyshev low-pass filtering is suitable, then, may know by the MAX274 component handbook: Regarding each filter unit, has:
In application, we MAX274 FC foot earth, then Rx/Ry=0.2, may calculate each resistance value from this. In practical application, because the resistance number is special, each resistance value must obtain by two precision resistor series.
1.5 DSP parts
This part is the block inserts on the ISA main line’s half long board, it controls the AD/DA part to complete the pronunciation data gathering and playbacking, and according to downloads the software to complete the corresponding signal analysis, is this system’s key component. It by TMS320C50 (operating frequency 40MHz), twin port RAM (4K×8), FIFO (1K×8) and the corresponding control circuit is composed.
When use, the DSP procedure downloads from the main engine through twin port RAM to the C50 internal procedure area. Before the sampling, C50 establishes the AD/DA part’s sampling frequency through internal timer Timer. After the single sampling point sampling completes, the AD/DA part produces C50 the hardware interrupt INT1, C50 response to interrupt INT1, will come from AD the sampled data to read in FIFO, the main engine through reads FIFO to obtain the sampled data; Before pronunciation playbacking, C50 establishes internal timer Timer according to the playbacking frequency, causes it to produce interrupts INT1; When playbacking, the main engine reads in the data FIFO, C50 fixed time to respond interrupts INT1, and delivers from the FIFO read data it the DA output; The signal processing duty completes in the C50 interior, the single processing input output data exchange through twin port RAM between the main engine and the DSP part.
To compile the software conveniently, in this we list the DSP part and PC machine as well as between the AD/DA part’s connection parameter:
Regarding PC
· the twin port RAM address range is D800H (segmented address): 000H~0FFFH (offset address).
· the order mouth CMD address is 300H, the order character design is as follows:
BIT0: Repositions FIFO:BIT1, BIT7: Has not used temporarily; BIT2: Repositions TMS320C50; BIT3:PC and C50 handshake signal: BIT4~6: Produces C50 to interrupt INT2~INT4.
· the condition mouth STS address is 0301H, the condition character design is as follows:
BIT0:FIFO2 spatial symbol; BIT1:FIFO1 full symbol; BIT2:FIFO1 half-full symbol; BIT3:C50 and PC handshake signal; BIT4~6 has not used temporarily.
· FIFO1 writes the address is 302H, reads the address is 303H.
Regarding DSP
· the twin port RAM address range is F000H~FFFFH.
· the condition mouth STS address is PA1, the condition character is as follows:
BIT5:FIFO1 spatial symbol; BIT6:FIFO2 half-full symbol; BIT7:FIFO2 full symbol; Other have not used temporarily.
· FIFO2 writes the address is PA2, reads the address is PA3.
· The AD sampling’s address is PA8.
· DA uses two levels of locks to save, the first level of lock saves the address is PA11, the second level of lock saves the address is PA10.
1.6 AD/DA parts
This part is inserts together in PC machine on the ISA main line’s half long board, its highest sampling frequency reaches 100kHz, under its DSP part’s control completes the pronunciation data the sampling and playbacking. This part’s input output range for ±10V, the resolution is 16bit. A/D transforms the data for the two’s complement form, the data which D/A transforms for the displacement binary code form.
2 system work software flow
This voice signal processing system by real-time mode or interrupt mode work, their software flow and structure more or less the same. Here take interrupt mode as the example showing software flow, like Figure 2 and shown in Figure 3. The entire work flow mainly controls the flow and the DSP interrupt processing flow constitution by the main engine. And INT1 uses in the voice signal the sampling and playbacking; INT2 uses in the main engine to the DSP calling order operational factor; INT3 uses in the main engine transmitting the DSP procedure to DSP (to have the different DSP procedure regarding different analysis processing). Figure 3 (a) is the DSP master routine flow chart, Figure 3 (b) for the INT1 interrupt servicing flow chart, Figure 3 (c) is the INT2 interrupt processing flow, Figure 3 (d) for the INT3 interrupt servicing flow.
3 system debugging result
Carries on the electricity performance test independently to various parts, meets the technical specification requirements after completely, carries on jointing shake down testing to the overall system. After the system complete electricity performance index meets the requirements, also carried on the massive software testing, the result has satisfied the design requirements. At present, this system already moves successfully in “the words automatic diagnosis” the experiment, uses it to be possible to complete the voice signal the LPC analysis, different points FFT analysis, real-time spectral analysis and so on.
