Introduction

In a conventional digital system, audio material is captured, stored, transmitted and reproduced as a sequence of values, which correspond to the amplitude variations of an electric signal at discrete points in time. Our ability to extract meaningful information from this raw data through either hearing or visualization of the signal curve is however somewhat limited to emotional interpretation, which as one may expect, is extremely subjective.

Extensive studies have shown that first converting this data to a so-called frequency representation is extremely useful for a broad range of audio applications, as it is quite similar in principle to the human auditory system. A proper detailed explanation of the reasons behind this is well outside of the scope of this manual, so we will only hint at a few important characteristics of human hearing, namely its

• ability to recognize and isolate sounds base on their relative intensity or loudness
• ability to identify a pitch and timbre (color, texture) for sounds that fall in this category
• ability to distinguish sounds based on their actual or perceived location

A number of methods have been invented in order to translate these characteristics to measurable quantities that can be expressed in standardized units. These provide invaluable tools for assessing the quality of an audio recording, assisting the engineer in detecting potential mix problems, conforming to industry standards and requirements, calibrating loudspeaker systems, tuning room acoustics, etc.

A fundamental tool for transforming a time-based digital audio signal into a frequency-based representation, a.k.a frequency spectrum, is the discrete Fourier transform (DFT) and its derivatives, such as the Short-Term Fourier Transform (STFT) and Fast Fourier Transform (FFT). Basically, the DFT maps a signal to a set of amplitudes taken at equally-spaced frequency intervals. In essence, one can see the DFT as a bank of many band-pass filters, with as many meters at the output of these filters.

Whilst constraining the frequencies to be taken at fixed, regular intervals, is convenient both in terms of processing resources and simplicity of the computation, amongst other reasons, this linear frequency binning does not represent human hearing, which is essentially logarithmic (constant Q), very accurately. The analysis engine in FLUX:: Analyzer therefore offers both options, which are discussed in more detail in Spectrum analyzer.