Content-based audio classification and retrieval using the nearest feature line method

Li, S.Z.

doi:10.1109/89.861383

Cited by 176 publications

(106 citation statements)

References 9 publications

Supporting

Mentioning

103

Contrasting

Unclassified

Order By: Relevance

“…The 409 sounds are partitioned into a training set of 211 sounds and a test set of 198 sounds, as that in [8]. The partition is done in the following way: (1) sort the sounds in each class in the alphabetical order of the file names, and then (2) construct the two sets by including sounds 1, 3, T V T U T in the prototype set and sounds 2, 4,…”

Section: Methodsmentioning

confidence: 99%

“…For CepsL feature sets, we set L=40 for the AdaBoost, because Ceps40 is relatively better than other S values based on our previous experiments [8] [9]. The boosting process is shown in the middle graph of Fig.…”

Section: T V T U Tmentioning

confidence: 99%

“…In Pfeiffer etal [12], audio features are extracted by using gammaphone filters. Li [8] used the nearest feature line (NFL) method for content-based audio classification and retrieval. Li and Guo [9] proposed to use the support vector machines (SVMs) to classify and retrieve audio patterns.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Boosting for content-based audio classification and retrieval: an evaluation

Guo

Zhang²,

Li³

2001

IEEE International Conference on Multimedia and Expo, 2001. ICME 2001.

Self Cite

View full text Add to dashboard Cite

In this paper, we evaluate a recently proposed algorithm in machine learning called AdaBoost for content-based audio classification and retrieval. AdaBoost is a kind of large margin classifiers and is efficient for on-line learning. Our focus is to evaluate its classification and retrieval accuracy as compared with other methods. The Muscle Fish audio database of 409 sounds is used for the evaluation with perceptual and cepstral features.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: T V T U Tmentioning

confidence: 99%

See 1 more Smart Citation

Boosting for content-based audio classification and retrieval: an evaluation

Guo

Zhang²,

Li³

2001

IEEE International Conference on Multimedia and Expo, 2001. ICME 2001.

Self Cite

View full text Add to dashboard Cite

show abstract

“…The basic idea underlying the NFL approach is to utilize all the possible lines consisting of any pair of feature vectors (prototypes) in a given training set to encode the feature space in terms of the ensemble characteristics and the geometric relationship. As a simple yet effective algorithm, the NFL has shown good performance in face recognition (Li, 1998), audio classification and retrieval (Li, 2000), and image classification . The NFL takes advantage of both the ensemble and the geometric features of samples for pattern classification.…”

Section: Introductionmentioning

confidence: 99%

On the use of nearest feature line for speaker identification

Chen

Zhang

2002

Pattern Recognition Letters

View full text Add to dashboard Cite

As a new pattern classification method, nearest feature line (NFL) provides an effective way to tackle the sort of pattern recognition problems where only limited data are available for training. In this paper, we explore the use of NFL for speaker identification in terms of limited data and examine how the NFL performs in such a vexing problem of various mismatches between training and test. In order to speed up NFL in decision-making, we propose an alternative method for similarity measure. We have applied the improved NFL to speaker identification of different operating modes. Its text-dependent performance is better than the dynamic time warping (DTW) on the Ti46 corpus, while its computational load is much lower than that of DTW. Moreover, we propose an utterance partitioning strategy used in the NFL for better performance. For the text-independent mode, we employ the NFL to be a new similarity measure in vector quantization (VQ), which causes the VQ to perform better on the KING corpus. Some computational issues on the NFL are also discussed in this paper.

show abstract

“…The second set of features consists of features which have been used by other groups for specific, complex recognition problems, in particular, speech recognition [9,10,13]. In the time domain the following 3 features were calculated: zero crossing rate, fluctuation of amplitude and 90%-10% width of amplitude histogram distribution.…”

Section: Sound Classificationmentioning

confidence: 99%