DSP Speech Processor Experimentation 2012

Version 1 - obsolete - see version 2.1 -

A DSP based speech processor based on the Atmel Mega88 microcontroller, designed to maximize voice levels on communications transceivers.

Now sounding even better!

Running on a 20MHz 8-bit AVR microprocessor, the unit works by writing and reading 16 bit audio data through a 256 byte circular buffer, whilst maximizing the waveform as a background task in-between samples. The peak level of each half-cycle of waveform is measured, and a multiplication factor is calculated and applied. Since the gain is updated at the waveform zero-crossings, distortion is minimized. The processed output sounds exactly like an SSB speech clipper, since it is higher in intermod, but low in harmonics. A sinewave will pass through undistorted.

Another function provided by the software is DDS generation of modulated audio frequency tones, for linearity testing.

The unit also makes some interesting guitar distortion: The notes beat against each other without the harsh distortion of audio clipping.

Yet to add:

Analytical processing - *done* The new software uses an efficient Hilbert Transform filter that splits the signals into "I" and "Q" channels
Phase rotation - *done*The Hilbert Transform filter also provides phase rotation as a nice side-effect!
EQ - *done* A simple 1st order IIR filter was used to reduce bass and boost treble
Filtering - Split band processing is being investigated as a way to improve audio quality
A processor upgrade to a dsPIC33 is scheduled for the near future, and will increase DSP power substantially
Receive audio processing, for digital muting/NR/filtering etc.