Signal conditioning

 

Signal conditioning includes all signal modifications aimed at optimising signal characteristics for special purposes. Examples of such purposes are presented in the next sections.

Modification of speech parameters

 

Modification of speech signals may intentionally influence some of the following parameters:

• The amplitude-frequency spectrum   of a speech signal can be altered by digital filters such as low-pass, high-pass or other types of filter. Other methods modify the speech characteristics parametrically: the most common coding scheme describes the strength and frequency of the time-variant vocal tract resonances called formants.
• Prosodic  characteristics like speaking rate  and fundamental frequency   may be modified independently in the time domain by applying a PSOLA algorithm   [Moulines & Charpentier (1990)]. Most PSOLA algorithms need speech signals labelled  with pitch  marks (fundamental period  pointers).
• The sampling rate  can be converted by using an algorithm for sample frequency  conversion. In decreasing the sampling rate  the upper cut-off frequency  and signal bandwidth  also have to be adopted.
• Other signal modification procedures modify not primarily features of the speech signal itself but channel  or environmental  characteristics and are mentioned below.

Signal contamination

   

Signal contamination aims at generating ``non-flawless'' speech signals under well-defined conditions. Starting with ideal ``flawless'' speech , a wide range of signal and channel  characteristics is subsequently adjustable by controlled speech contamination. The ``flaws'' can be chosen from a variety of linear or non-linear distortions (cf. Section 8.6). Possible applications are the stimuli generation process for the evaluation of the robustness of speech recognisers  or the assessment of speech coding algorithms under non-ideal conditions.  

Binaural processing

   

Binaural processing simulates the signal processing scheme of the human auditory system [Blauert (1983)]. By investigating the relevant signal parameters from both ears it is possible to make full use of signal features that are impossible to obtain from monophonic signals. These features include positions, movements and extensions of sound sources. In the field of speech, interesting applications for binaural techniques are speech enhancement and voice separation, for example in connection with the so-called cocktail party effect, i.e. the ability to track one voice among many.  

Simulation of environmental factors

   

Special algorithms like room simulation or reverberation  tools allow subsequent introduction of acoustic environmental  characteristics into the speech signal (cf. Section 8.5 on environment characteristics). Beyond impressions like room size and wall properties, sophisticated virtual auditory environments  can be simulated in combination with binaural processing  techniques. Although the speech signal has been produced independently of these environmental factors, arbitrary room acoustics can be created from a single recording and simulated without the physical existence of the room.