Discovering hierarchical speech features using convolutional non-negative matrix factorization
By Sven Behnke. 2003
49 citations recorded by the source corpus.
Discovering a representation that reflects the structure of a dataset is a first step for many inference and learning methods. This paper aims at finding a hierarchy of localized speech features that can be interpreted as parts. Non-negative matrix factorization (NMF) has been proposed recently for the discovery of parts-based localized additive representations. The author proposes a variant of this method, convolutional NMF, that enforces a particular local connectivity with shared weights. Analysis starts from a spectrogram. The hidden representations produced by convolutional NMF are input to the same analysis method at the next higher level. Repeated application of convolutional NMF yields a sequence of increasingly abstract representations. These speech representations are parts-based, where complex higher-level parts are defined in terms of less complex lower-level ones.
Subfields: Artificial Intelligence
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- J. Urgen Schmidhuber
- Karthik Devarajan
- David R Ha
- Alex Graves
- Sergey Levine
- Yoshua Bengio
- Quoc V. Le
- Faustino J. Gomez
Citations in the corpus
- Deep learning in neural networks: An overview (2014, methodology): “See also more recent methods using UL to improve subsequent SL (e.g., Behnke, 1999, 2003a; Escalante-B & Wiskott, 2013).”
- Nonnegative Matrix Factorization: An Analytical and Interpretive Tool in Computational Biology (2008, background): “…and facial pattern recognition [8–19], natural language processing such as in text mining and document clustering (see [20–22] and references therein), sparse coding [23–27], information retrieval [28,29], speech recognition [30–33], video summarization [34], and Internet research [35,36].”
- Generalized Nonnegative Matrix Approximations with Bregman Divergences (2005, unknown): “An application to the discovery of hierarchical speech features appears in a paper by Behnke [2].”