Discussion - Intelligent Audio Analysis - page 308

Chapter 13
Discussion
A scientist's aim in a discussion with his colleagues is not to
persuade, but to clarify.
— Leo Szilard
Picking up on the aims of this topic as were outlined in Sect. 2.2 , these are discussed
one by one as follows. This includes a statement on how the state-of-the-art in
the field was advanced more recently. Basing on these, a distilled 'best practice'
recommendation is given to the reader, before a critical discussion of missing aspects
and remaining research steps.
13.1 Picking Up on the Goals
(I) A unified perspective on audio analysis tasks was provided in this topic in the
hope to bridge between the disciplines of speech and language processing—including
sub-disciplines such as ASR, computational paralinguistics or opinion mining and
sentiment analysis—music analysis, i.e., MIR, and general sound analysis. Even
though these often co-exist side-by-side and transfer is often limited, the current
book by intention provided a unified perspective to reach first and foster future
synergies. Further, a broad overviewwas given on recent advancements in these fields
by presentation of manifold exemplary applications and performances in realistic
conditions. In particular, realistic conditions were mostly ignored in the field so far
by a number of simplifications.
(II) Recent methods were presented and shown in application. These aim to
improve robustness and reliability of today's Intelligent Audio Analysis systems. In
particular, the enhancement and isolation of the signal of interest by suited methods
of blind separation of sources was emphasised on [ 1 - 9 ]. This is sparsely found in the
field at this time coupled with subsequent 'intelligent' processing. Instead, research
usually focuses either on the separation or on intelligent processing of 'clean' and