SALIENCY DEGREE ESTIMATION METHOD, SALIENCY DEGREE ESTIMATION DEVICE, AND PROGRAM

• Main Authors: KAYANO MAKIO, YONEYA MAKOTO, FURUKAWA SHIGEHITO
• Format: Patent
• Language: English; Japanese
• Published: 25.07.2016
• Subjects: ACOUSTICS; AIDS FOR MUSIC; CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; MUSICAL INSTRUMENTS; OTHER AUXILIARY DEVICES OR ACCESSORIES FOR MUSIC OR MUSICALINSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION; SUPPORTS FOR MUSICAL INSTRUMENTS

PROBLEM TO BE SOLVED: To provide a saliency degree estimation method capable of estimating the saliency degree of objective sound based on unpredictability of a time series pattern.SOLUTION: A saliency degree estimation method estimates the saliency degree of sound in part of the time section of an input signal and includes: a similar partial section detection step for detecting one or more similar partial sections that is a partial section similar to a reference signal from within a storage interval; a prediction distribution generation step for generating one or more prediction distributions that is a distribution based on the feature quantity of an input signal corresponding to a predetermined time section just after the detected similar partial section; and a saliency degree estimation step for estimating the saliency degree of an input signal corresponding to a prediction interval based on the generated prediction distribution and an estimation object signal.SELECTED DRAWING: Figure 1 【課題】時系列的なパターンの予測不可能性に基づく対象音の目立ち度合いを推定できる顕著度推定方法を提供する。【解決手段】入力信号の一部の時間区間の音の目立ち度合いを推定する顕著度推定方法であって、参照信号と類似する部分区間である類似部分区間を蓄積区間内から１つ以上検出する類似部分区間検出ステップと、検出された類似部分区間の直後にある所定の時間区間に対応する入力信号の特徴量に基づく分布である予測分布を１つ以上生成する予測分布生成ステップと、生成された予測分布と推定対象信号に基づいて、予測区間に対応する入力信号の顕著度を推定する顕著度推定ステップを含む。【選択図】図１