Acoustic Lexemes for Organizing Internet Audio

In this article, a method is proposed for automatic fine-scale audio description that draws inspiration from ontological sound description methods such as Shaeffer's Objets Sonores and Smalley's Spectromorphology. The goal is complete automation of audio description at the level of sound objects for indexing and retrieving sound segments within Internet audio documents. To automatically segment audio documents into acoustic lexemes, a hidden Markov model is employed. It is demonstrated that the symbol stream of cluster labels, generated by the Viterbi algorithm, constitutes a detailed description of audio as a sequence of spectral archetypes. The ASCII base-64 encoding scheme maps cluster indices to one-character symbols that are segmented into 8-gram sequences for indexing in a relational database. To illustrate the methods, the essential components of an audio search engine are described: the automatic cataloguer, the retrieval engine and the query language. The results of experiments that test the accuracy and the retrieval efficiency of six new similarity-matching algorithms for audio using acoustic lexemes are presented. The article concludes with examples of audio matching using the structured query language (SQL) for creating new musical sequences from large extant audio collections.