Tetsuya Muroi

Speech segment detection and word recognition

In a speech segment detection method, a sequence of speech samples is provided from an input speech signal and a sequence of feature vectors is provided from the speech samples, the feature vectors having respective speech power levels. A minimum speech power among the speech power levels in the feature vector sequence is detected. Normalized speech power levels are computed based on the speech power levels and the minimum speech power. Each of the normalized speech power levels is compared with a predetermined threshold value to detect speech segments in the input speech signal. Further, a speech recognition system and method and a computer-readable medium, using the speech segment detection method, are also disclosed.

Voice recognition system using voice power patterns

A voice processing method and system using voice power is provided. A voice power signal is produced in analog format and this voice power signal is sampled at a predetermined time interval to determine its current envelope or amplitude, which is then compared with a predetermined number of voice power levels, whereby 1 is assigned to at least one of the predetermined number of voice power levels which corresponds in level to the amplitude thus determined and 0 is assigned to the rest of the predetermined number of voice power levels, thereby converting the analog voice power signal into a binary voice power pattern. In the preferred embodiment, the voice signal is also processed through a frequency analyzer including a plurality of band-pass filters different in frequency range and a binary converter to form a binary time-frequency voice distribution pattern which is then combined with the binary voice power pattern to form a combined voice pattern. As a modification, a voice power range is determined by multiplying a predetermined threshold to the amplitude, and the voice power pattern is compared with the predetermined number of voice power levels.

Speech recognition method and apparatus

A reference speech pattern is described as a time series of a fixed number of states. Each of states is described with one feature vector (centroid) which represents a feature quantity of a speech portion contained in the state, and a duration time of that state. Pattern matching is carried out between the time series of feature vectors of the input speech pattern and the time series of the states which describe the reference pattern. A weighting function modifies the pattern matching distance depending on differences in duration times between the input speech and the reference pattern.