Audio-based gender identification using bootstrapping

Annotation of audio content is an important component of modern multimedia information retrieval systems. Automatic gender identification is used for video indexing and can improve speech recognition results by using gender-specific classifiers. Gender identification in large datasets is difficult because of the large variability in speaker characteristics. Bootstrapping is an approach that attempts to combine minimal user annotations with automatic techniques for audio classification. In bootstrapping a small random sampling of the training data is annotated by the user and this annotation is used to train a classifier that annotates the remaining data. This technique is useful when the training set is too large to be fully annotated by the user. Experimental results showing that bootstrapping is effective for automatic audio-based gender identification are provided.