Access count of this item: 177

Files in This Item:
File Description SizeFormat 
ieice_tec.rep_SP2017-31.pdf1.41 MBAdobe PDFView/Open
Title: 雑音・残響下におけるRahmonicとメルケプストラムを用いた叫び声検出
Other Titles: Detection of noisy-and-reverberant shouted speech using rahmonic and mel-frequency cepstrum coefficients
Authors: 福森, 隆寛  KAKEN_name
中山, 雅人  KAKEN_name
西浦, 敬信  KAKEN_name
南條, 浩輝  kyouindb  KAKEN_id
Author's alias: Fukumori, Takahiro
Nakayama, Masato
Nishiura, Takanobu
Nanjo, Hiroaki
Keywords: 叫び声検出
Shouted speech detection
Noisy-and-reverberant environment
Mel-frequency cepstrum coefficients
Issue Date: Aug-2017
Publisher: 電子情報通信学会
Journal title: 電子情報通信学会技術研究報告 = IEICE technical report : 信学技報
Volume: 117
Issue: 189
Start page: 49
End page: 54
Thesis number: SP2017-31
Abstract: 本稿では, 雑音・残響環境下におけるRahmonicとメルケプストラム(Mel-Frequency Cepstrum Coefficients: MFCCs)を用いた叫び声検出手法について述べる. 人間の聴覚特性を考慮したケプストラム係数であるMFCCsは, 音韻を特定するための声道特徴量を示しており, また基本周波数の低調波成分であるRahmonicは, 人間の声帯運動に関わる特徴を表現している. これまで, 我々は大量の平静音声と叫び声から抽出したMFCCsとRahmonicに基づいて構築した3種類の音響モデル(GMM: Gaussian Mixture Model, HMM: Hidden Markov Model, DNN: Deep Neural Network)を用いて叫び声検出手法の有効性を示していた. 特に前報までは, クリーン環境と雑音環境における叫び声の検出性能を評価し, 提案手法が高い検出性能を確認した. 本稿では, 更に実環境を想定して, 雑音と残響が混在する環境において叫ばれた音声の検出性能を評価する. 評価実験の結果, MFCCsとRahmonicを音声特徴量として用いることで, 雑音や残響の種類やSNRによらず, 叫び声の発声機構(声道特性と声帯特性)を効率よく表現できることを確認した. また, ほとんどの騒音・雑音環境において音響モデルとしてDNNを用いることでGMMやHMMよりも高い叫び声検出性能を達成できた.
This paper describes a method based on new combined features with mel-frequency cepstrum coefficients (MFCCs) and rahmonic in order to robustly detect a shouted speech in noisy-and-reverberant environments. MFCCs collectively make up mel-frequency cepstrum, and rahmonic shows a subharmonic of fundamental frequency in the cepstrum domain. In our previous method, Gaussian mixture models (GMM), hidden Markov model (HMM) and deep neural network (DNN) are constructed with the proposed features extracted from training data which includes a lot of normal and shouted speech samples. Especially, our latest study showed the effectiveness of our proposed method through detection experiments for shouted speech in clean and noisy environments. In this study, we evaluate the detection performance of shouted speech in noisy-and-reverberant environments. The results show that MFCCs and rahmonic were effective for representing an utterance mechanism including both vocal tract and vocal cords, and these features were independent of noise and reverberation. In addition, DNN could achieve higher performance in noisy-and-reverberant environments than GMM and HMM.
Description: オーガナイズドセッション「音の認知・知覚機能の情報処理」(一般公演)
Rights: copyright © 2017 IEICE
Related Link:
Appears in Collections:Journal Articles

Show full item record

Export to RefWorks

Export Format: 

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.