The growth of DSP processors for professional audio processing (and some domestic applications) has opened up a number of exciting improvements for sound systems, in particular the possibility of correcting loudspeaker frequency responses using FIR filters.
FIR filters are able to equalise the frequency response of any audio signal (including phase and time alignment), typically deploying 256 to 1000 data point correction over the bandwidth of the device. This produces a far more accurate response that has been available until now.
measured response of an FIR correction filter applied to a horn loudspeaker - before (red) and after (blue)
FIR filters are typically generated using proprietary software in laboratory environments, and are usually only available for commercial products, having been generated by that company's engineers. David has developed his own process and can generate accurate FIR filters for loudspeaker correction not only in the laboratory, but also in the field.
• Sydney Trains have deployed one of his filters to dramatically improve the sound quality of cheap horn speakers installed on railway platforms.
• The Sydney Opera House are utilising a number of loudspeakers running his filters that were generated both in the laboratory and on site.
There has been a rapid expansion of delivering audio programs online such as podcasts and on-demand radio services. While the productivity has thrived, there are some areas of audio quality that have been left behind in the rush to deliver the online experience.
In former times, the master processing was a multiband compressor/limiter and an equaliser in the signal chain before the transmitter. Now the on-demand programs are one of a thousand files residing on a server, being accessed by many people at any one time.
The levels throughout a podcast can vary noticeably, particularly when interviews or music is included. Also, when low bit rates are used, sibilance can be exacerbated and an annoying and debilitating 'essing' sound can permeate the podcast.
David has developed a batch processing solution that adjusts all audio files so that the perceived sound level is constant irrespective of how the programs various sections were recorded (often by multiple people with different personal standards), the recording media, whether music is part of the program or not, different genders of announcer, stings from other departments, etc.
This processing does one or all of the following:
• adjusts level to ITU BS1770-2 standards
• applies audio compression with minimal 'hunting' or 'breathing'
• removes sibilance from audio without altering all the other program material
• applies filters and equalisation as desired
The result is:
• all programs are at the same perceived level
• consistent audio throughout the program : the customer doesn't have to turn up the soft passages or grab for the volume control at a loud burst of sound
• removal of sibilance from badly processed files or low bit rate anomalies
• adjust the frequency spectrum
This can all be applied automatically offline.
Sydney Trains: batch processing of thousands of audio files for the Digital Voice Announcement System to concatenate words into sentences.
The process imports sound files recorded:
• at different levels,
• by different announcers,
• of different genders,
• using differing recording methods,
and processors the files so that they have equal perceived level, removes sibilance, then exports them as sound files of a user selectable format.
The FINALISED level is absolute, with different recording sessions producing equal sound levels over the years.
Read most sound files from a hard drive, process these files, then write them back in different formats to multiple folders.
READ: Input File Types
WAVE (.wav) OGG (.ogg) FLAC (.flac) AU (.au) AIFF (.aiff, .aif) AIFC (.aifc) MP3 (.mp3) MPEG-4 AAC (.m4a, .mp4)
WRITE: Output File Types
WAVE (.wav) OGG (.ogg) FLAC (.flac) MPEG-4 AAC (.m4a, .mp4)
sample frequency & bit rate conversion - channel summation/splitting - filtering - compression - levelling