Abstract: This article proposed based on the TI Corporation’s TMS320VC5402 Chinese speech synthesis system plan. The overall system as the core electric circuit carries on the design take TMS320VC5402, uses take the syllable as unit’s neutral intonation, coordinates to overlap the additive PSOLA algorithm based on the tone synchronization, and (CELP) the encoding method carries on the code compression with the code drive linear prediction to the primitive sampling sound storehouse.
Key word: DSP; SCM; PSOLA; Text analysis; Speech synthesis
0 introductions
Develops unceasingly along with the voice signal processing technology and mature, the speech synthesis is becoming in gradually the information technology the man-machine connection key technologies. The DSP chip, namely the digital signal processor, is to realize fast specially, has the special structure microprocessor which each kind of signal processing algorithm designs, its processing speed compared to quickest CPU also quick 10~50 times. This article introduces is one kind realizes the method based on the DSP Chinese speech synthesis system.
1 system overall concept
Speech synthesis’s most major characteristic is must synthesize the infinite dictionary from the limited memory cell the continued statement to come [1]. In order to achieve this point, this system design by (1) front end has pretreated the standard format which the module the enter text file conversion system might process; (2) the rhythm regular storehouse gives under the current language environment each syllable rhythm characteristic parameter; (3) the voder basis assigns the rhythm characteristic parameter the corresponding pronunciation unit’s acoustics parameter makes the adjustment to the primitive pronunciation storehouse in; (4) after the adjustment each pronunciation unit will splice in together, obtains correspondingly in the enter text continuous speech output and so on 4 basic flow compositions. System basic principle diagram as shown in Figure 1.
2 hardware system design
* funding project: Hunan Province education department (03C025)
This Chinese speech synthesis system’s goal is to the input text document can, be able to understand naturally clearly, by the continuous speech form output. ATMEL Corporation’s AT89S52 monolithic integrated circuit carries on the demonstration to keyboard entry’s text document, delivers the TMS320VC5402 processing again, finally will synthesize the result output. Hardware architecture diagram as shown in Figure 2.
2.1 keyboard electric circuits and display circuit
At89S52 keyboard interface electric circuit is uses the interrupt mode, has when the key presses down, has the interrupt request, enters the interrupt processing, then through inquires P1.0 and the P1.1 situation, carries on corresponding processing. Sufficed with the resistance and the electric capacity to disappear shake the electric circuit, prevented electric circuit’s misoperation.
AT89S52 gives from keyboard entry’s text information LCD to demonstrate that simultaneously writes about the data exterior memory CY7C133, then TMS320 VC5402 reads on CY7C133 the information to carry on processing. LCD instruction format as shown in Table 1:
Table 1 LCD instruction format
|
RS R/W |
DB7 DB6 DB5 DB4 DB3 DB2 DB1 DB0 |
And RS, R/W decided together which register chooses, as shown in Table 2:
Table 2 register choices
|
RS |
R/W |
Operation |
|
0 |
0 |
The instruction register reads in |
|
0 |
1 |
Busy symbol and address |
|
1 |
0 |
The data register reads in |
|
1 |
1 |
Data register read-out |
2.2 TMS320VC5402 and AT89S52 correspondences
AT89S52 and the TMS320VC5402 respective independent working, its information and the data exchange through share a piece of exterior memory to realize, between them the signal contact passes the perfect connection and the software judgment realizes [2].
What the exterior memory uses is CY7C133, it is high speed 2K X 16bit static asynchronous twin port RAM, its storing velocity is 25ns. It has two sets of independent address wires, the data line and the control holding wire, allows in two control component’s data through the memory which connects together to carry on the correspondence. This twin port RAM allows two controllers simultaneously to read any memory cell (including simultaneously to read identical unit), as soon as soon as but does not allow simultaneously to write or reads writes the unified address unit.
Regarding TMS320VC5402, the data-carrier storage CY7C133 corresponding address is 4000H~47FFH.
Regarding AT89S52, the data-carrier storage CY7C133 corresponding address is 2000H~27FFH.
3 software system design
Chinese takes one kind to have the accent language, its rhythm characteristic is complex. For synthesizes the infinite dictionary from the limited memory cell the continued statement to come, must make the adjustment under certain rhythm rule to the pronunciation storehouse unit rhythm parameter, by obtains conforms to current language class environment phonetic change unit [3].
According to obtains the phonetic change unit method to be different, may divide into the voder two types: (1) profile splicing synthesis; (2) the parameter synthesis (calls source/filter synthesis). This system uses the profile splicing synthesis the method directly makes the adjustment to the profile time domain and the frequency range profile by the phonetic change unit which obtains needs.
The simple profile splicing law is very difficult to carry on the pitch and the length (time length) adjusts. Therefore, this system uses the neutral intonation syllable direct splicing coordination tone synchronization profile superimposition (PSOLA) the algorithm, and (CELP) the encoding method carries on the code compression with the code drive linear prediction to the primitive sampling sound storehouse the method. Basic flow chart as shown in Figure 3.
4 concluding remark
This system uses the SCM real time display input the text document, can synthesize the pronunciation and the enter text document realizes the contrast output, intuitive; Has the clarity, to be able understanding and the nature is high; The synthesis algorithm operation order of complexity is low, can exhaust the quantity small sound storehouse to realize to the capacity limited storage space takes the degree the request.
Reference:
[1] wears the recluse, Liang Xiaowen, the Pei small equality. Based on DSP modern electron system design [M]. Beijing: Electronics industry publishing house .2002
[2] Yi Kechu, Tian Bin, pays. Voice signal processing. Beijing: Defense industry Publishing house, 2000
[3] Huang Haibo, Jiang Weirong, Cheng Dengliang. General pronunciation processing system’s DSP realizes the [J] micro computer information 2006,5,173-175
