Parallel searching method for speech recognition
A speech recognition and search method technology, applied in speech recognition, speech analysis, instruments, etc., can solve the problems that hardware resources cannot be fully utilized, the load balance of processor calculations cannot be effectively controlled, and the distribution of search paths cannot be guaranteed. Achieve low communication overhead and realize the effect of computing load balance
Inactive Publication Date: 2005-05-11
INST OF ACOUSTICS CHINESE ACAD OF SCI +1
View PDF1 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
In this way, since the generation of the local search path depends on the voice input, the algorithm cannot guarantee that the distribution of the search path on each sub-search network is relatively uniform, so it cannot effectively control the load balance of the calculation load between the processors, making the calculation amount The processor with the highest load becomes the bottleneck of the system, while other processors are in a waiting state at this time, so the hardware resources of the system cannot be fully utilized
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0033] In the process of speech recognition, the calculation amount of token expansion (including Gaussian probability calculation of speech frame) accounts for more than 80% of the calculation amount of the entire search engine. In the environment of multi-processors or multi-core processors, in order to speed up the search operation, the present invention proposes a method for performing parallel operation on token expansion by using multi-threads.
[0034] There is data dependency in the token extension process in the token chain list of multiple states, and it still ends with Figure 4 As an example, assume that the active token linked lists of states i2, i3 and j1 are H i2 、H i3 and H j1 , at time t-1, there are two active tokens T in each of the above token linked lists 11 (t-1)-T 12 (t-1), T 21 (t-1)-T 22 (t-1) and T 31 (t-1)-T 32 (t-1), at time t, it is necessary to expand the active tokens in the three state token lists respectively, and the expansion order of...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
This invention discloses a parallel searching method in phonetic indentification, in which, multiple threads carry out parallel operations to token expansion, everyone gets its own t-1 time scale dictionary tree token resources and to the assigned t-1 token resources, thread Ti engages in expansion operation to otkens in all station token chain lists at each state and puts the new generated token chain in the token chain list corresponding to Ti in target state, after finishing the expansion operations, symmchronous operations of token resources are carried out realizing computing load balance among processors in a multiprocessor system.
Description
technical field [0001] The invention relates to a speech recognition method, in particular to a parallel search method based on dictionary tree width priority strategy in speech recognition. Background technique [0002] The voice server is the main component of the distributed pronunciation recognition system and the voice portal system, and the search engine occupies more than 80% of the CPU computing resources and most of the memory in the entire voice server. [0003] figure 1 Shown is a structural block diagram of a known speech recognition system. The analog speech is converted into a computer-processable digital signal through an analog-to-digital conversion unit 11, and then the digital signal is processed by a frame by a feature extraction unit 12. Usually, the frame length is is 20ms, frame shift is 10ms, extracts the MFCC parameter of each frame of speech, obtains the MFCC vector sequence, decoding operation unit 14 adopts certain search strategy according to the...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Patents(China)
IPC IPC(8): G10L15/32
Inventor 韩疆庹凌云潘接林刘建颜永红王俊杰
Owner INST OF ACOUSTICS CHINESE ACAD OF SCI