초록 ▼

A speech recognizing device capable of recognizing a speech uttered in a noisy environment. The device uses low frequency data which are hardly susceptible to spectrum fluctuations for those portions of an input speech which are similar to vowels, while using entire frequency data for the other port

A speech recognizing device capable of recognizing a speech uttered in a noisy environment. The device uses low frequency data which are hardly susceptible to spectrum fluctuations for those portions of an input speech which are similar to vowels, while using entire frequency data for the other portions of the speech.

대표청구항 ▼

A speech recognizing device comprising: entire frequency analyzing means for producing an entire frequency input pattern which is a sequence of entire frequency characteristic vectors representative of data associated with an entire frequency band of an input speech; low frequency analyzing means fo

A speech recognizing device comprising: entire frequency analyzing means for producing an entire frequency input pattern which is a sequence of entire frequency characteristic vectors representative of data associated with an entire frequency band of an input speech; low frequency analyzing means for producing a low frequency input pattern which is a sequence of low frequency characteristic vectors representative of data associated with a low frequency band of the input speech; entire frequency standard pattern storing means for storing entire frequency standard patterns which are sequences of entire frequency characteristic vectors of predetermined standard speech; low frequency standard pattern storing means for storing low frequency standard patterns which are sequences of low frequency characteristic vectors of predetermined standard speech; weighting coefficient processing means for storing predetermined weighting coefficients; entire frequency distance calculating means for calculating an entire frequency distance between the entire frequency characteristic vectors of the entire frequency input pattern and the entire frequency characteristic vectors of the entire frequency standard pattern; low frequency distance calculating means for calculating a low frequency distance between the low frequency characteristic vectors of the low frequency input pattern and the low frequency characteristic vectors of the low frequency standard pattern; frame distance calculating means for determining a weighted distance by calculating a weighted summation of the entire frequency distance and the low frequency distance using the weighting coefficients; and recognizing means for recognizing the input speech by using the weighted distance.