Acoustic topic model for audio information retrieval

Acoustic scene classification based on generative model of acoustic spatial words for distributed microphone array

“…Focusing on this idea, some researchers have employed the BoW representation in other research fields such as computer vision (bag-of-visual words) or acoustics (bag-of-acoustic words [7], [14], [19], bag-of-angle words [23]). Specifically, the bag-of-acoustic words is a discrete feature representation that quantizes the spectral features of sounds into acoustic words and aggregates acoustic words into a histogram of them.…”

“…For instance, an acoustic scene "cooking" is characterized by a combination of multiple sound events including "running water," "cutting ingredients," and "heating a skillet." On the basis of this idea, Guo and Li [13], Kim et al [14], and Imoto and coworkers [7], [15] proposed acoustic scene classification methods based on the bag-of-acoustic words, which quantize the spectral features into acoustic words and aggregates acoustic words into a histogram of them.…”

Acoustic Scene Analysis Based on Hierarchical Generative Model of Acoustic Event Sequence

“…Acoustic scene estimation methods that focus on the latent structure of an acoustic scene lying in acoustic event sequences have also been proposed, and these are called the "acoustic topic model (ATM)" [11]- [13]. In the ATM, acoustic scenes are represented as a distribution of latent variables called "acoustic topics," and a generative process of acoustic event sequences e s is modeled not by using acoustic scenes directly but through distributions of acoustic topics as shown in Fig.…”

“…We call this model the "acoustic scene model (ASM)." Kim et al [11], Lee et al [12], and Imoto et al [13], [14] proposed another unsupervised generative model for analyzing acoustic scenes on the basis of the "acoustic topic model (ATM)." The ATM is an unsupervised generative model that generates acoustic event sequences from acoustic scenes through the latent structure of the acoustic scene lying in a combination of acoustic events, which is called the "acoustic topic."…”

Classification of Overlapped Audio Events Based on AT, PLSA, and the Combination of Them

“…So far there has been no report on applying AT in audio field, but much work has been done on applying LDA in audio retrieval. For example, Samuel Kim [14] assumed that an audio clip was a mixture of some acoustic topics, and took LDA to extract the topic distribution information for each audio clip to realize audio retrieval. Pengfei Hu [4] overcame the shortage of LDA in processing continuous data, and proposed a new topic model named Gaussian-LDA for audio retrieval.…”