SYSTEM FOR REGISTERING VOICE STANDARD PATTERN

• Main Authors: TODA KEIKO, MORITO MAKOTO
• Format: Patent
• Language: English
• Published: 02.05.1990
• Subjects: ACOUSTICS; MUSICAL INSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION

PURPOSE:To select a satisfactory voice pattern independent of a segmented result, to prepare a voice standard pattern with good accuracy and stability and to improve a recognition performance by correcting the vocalizing speed of the speaker of a standard vocalizing length and setting the normal value of the selection of the voice pattern. CONSTITUTION:A writing means 310 collects the voice pattern and vocalizing length of the voice standard pattern from information from a feature extracting part 14 and a voice segmenting part 16. Based on the mean vocalizing length read from a mean vocalizing length table 32 and the vocalizing length, a (vocalizing length ratio) vocalizing speed ratio is calculated, and it is written to a temporary storage device 34. A voice pattern selection normal setting means 320 reads the vocalizing length ratio and mean vocalizing length and obtains a maximum permissible vocalizing length and a minimum permissible vocalizing length by using them. Whether the length is included within the range of the permissible vocalizing length or not is discriminated, the corresponding voice pattern is read from the temporary storage device 34, and it is selected as a candidate pattern for preparing the voice standard pattern. Thus, the preparation of the voice standard pattern with good accuracy and stability can be executed.