Method for quickly forming voice data base for key word checkout task

Abstract

A method for quickly structuring voice databank used in key word detection task includes recording voice databank of isolated word, confirming key word table and nonkey word table according to requirement of key word detection system, confirming parameters such as total sentence number and time length as well as key word occurrence frequency for key word detection test, connecting key word and nonkey word in accordance with requirement of key word detection system to be sentence by pasting - up means for generating out databank used in key word detection task.

Description

technical field [0001] The invention relates to a construction method of a speech database, in particular to a construction method of a speech database used for keyword detection tasks. Background technique [0002] The fundamental purpose of speech recognition research is to realize the interaction between human beings and machines with natural language, so that the machine has the same auditory function as human beings, can directly receive human speech, understand human intentions and make corresponding responses. Speech is the most natural, most convenient, and most commonly used way of information exchange for human beings. Keyword detection is the identification of a given set of words—called keywords—from a continuous, unrestricted stream of natural speech. Keyword recognition is a branch and an important research direction of speech recognition. Keyword detection technology has demonstrated great value in many application systems. It has brought speech recognition ...

AI Technical Summary

Problems solved by technology

[0005] (1) It is difficult to record and organize the voice database

[0006] At present, the voice database used in the keyword detection system is actually recorded and collected under specific requirements. Each recording needs to involve many speakers and operators, and the workload of the voice database is very huge; The database performs post-processing such as labeling, and the workload is several times that of the recording process itself; at the same time, factors such as accent, spoken language, and noise are uncertain and uncontrollable, and the recording work is difficult and inefficient

[0007] (2) Poor flexibility of speech database

[0008] After each collection of the voice database, some characteristics of the database, such as the length of sentences and the distribution of the number of occurrences of keywords, are relatively fixed; even if a certain change can be achieved by selecting a subset of the database, such changes are Very limited, and the size of the database is also reduced, because once the voice data is collected, each word (whether defined as a keyword or a non-keyword) is fixed in the database

[0009] (3) The scale of the database is small, and it is not yet comprehensive for the performance of the inspection system

[0010] Due to the huge workload of actually collecting the voice database, the scale of the voice database used in the keyword detection system is relatively small; at the same time, due to the particularity of the test of the keyword detection system, the real "keyword" in the continuous voice stream The probability of occurrence is generally small, and the occurrence of many keywords cannot appear more times, thus affecting the effective test of system performance

[0011] (4) The characteristics of keyword appearance are uncontrollable

[0012] For the speech database of the test keyword system, it is generally hoped that its characteristics such as the frequency of keywords, the position of the keywords in the sentence, and the distribution of the number of times the keywords appear in each sentence meet certain requirements; It is extremely difficult to completely meet the preset settings in the word speech database, and once a specific setting is satisfied, it is impossible to change it

Images