While current GSM audio codecs are limited for 3.4kHz narrowband speech audio, the introduction of a 7kHz wideband speech audio provides an improved voice quality especially in terms of increased naturalness in voice. AMR (Adaptive Multi Rate) coding ]brings speech wireline quality to current GSM technology. There is a growing market interest in AMR speech coding among operators and service providers in order to provide highest quality of audio services while creating user awarness to the data services.

AudioBlaze™ can convert different sound formats into AMR specific audio files. Accepted sound formats are as follows:

Apple/SGI AIFF	Sun AU
NeXT .snd	PCM, U-law, A-law ADPCM
CD-R	AMR
Microsoft WAV	MP3
Psion A-law WVE	MIDI
Sound Blaster .VOC	GSM

AudioBlaze™ contains several audio processing implementation including Concatenation which can be used to add small advertisements at the end/begining of a sound file for GSM subscribers:

Volume Adjustment: Increases and decreases the volume of a sound source.
Cropping: Removes specified time intervals with in the source sound. Features including trimming off the silence from the beginning of a sound file.
Concatenation: Adds another sound stream into the sound source. Additional sounds can be added anyplace of the sound source (ie. From a time frame to the end, between time frames, beginning of the time frame ..etc)
Reversing sound file: This feature helps to create reverse playing of the sound object.
Changing the speed: This feature can increse and decrease the speed of the sound source. (ie: like speeding up the motor on a tape recorder)
Adding reverb effect: Adding reverb effect to a sound source.

There exist no limitation for the source file quality. AudioBlaze™ converts all common file formats. AudioBlaze™ downsamples the source file, if it is above quality standard compared to AMR. As an example, AudioBlaze™ can convert 44.1kHz, 16 bit, 2-channel stereo, 176.400Kb/s WAV file into respective AMR audio file without extra process.

Converted AMR files are stored as 4.8kbps, 8kHz sampled mono. If audio sources are high quality files, AudioBlaze™ performs downsampling process, in which signal energy is released. In Appendix-I , an experiment result regarding this process can be reviewed. This experiment also uses a mathematical signal source, random noise which has a flat spectrum at which contains constant energy at all frequencies, in order to better understand noise filtering enchancements performed by AudioBlaze™ algorithms. The signal source is sampled at 44.1kHz and written on a 16-bit single channel WAV. Results are observed by using MATLAB. The resulting spectrum is plotted in dB against the AMR frequency.

As can be seen from results, there is almost no aliasing but there exist some peak spreadings between 20-40dB lower corner of the main peak which is not perceivable by human. The overall result is almost a correct, fine signal as it is expected.