Mixed-Excitation Linear Predictive

from Wikipedia, the free encyclopedia

MELP ( M ixed- E xcitation L inear P redictive vocoder Algorithm ) is an audio codec included for lossy compression of audio data, the only human speech.

Compared to compression methods like MP3 or Vorbis , it is unsuitable for music data; it is designed for an intelligible reproduction of human speech at extremely low bit rates of 600, 1200 or 2400 bit / s. It delivers relatively good results with strong background noise and requires little power.

MELP is the US Department of Defense's standard method for digital voice transmission at 2400 b / s.

With the extremely low bit rates for which the codec is designed, acoustic enjoyment is hardly possible - if more bandwidth is available, Speex delivers better results from this point of view. The free narrowband speech codec Codec2 can also be seen as an alternative. This delivers at 1200 bit / s z. T. better results than MELP.

The codec is standardized as both MIL-STD-3005 (2400 bit / s) and STANAG -4591 (NATO). I.a. the 600 bit / s variant is called MELPe , with the e for ' enhanced ' (in German: 'improved').

Copyright Claims and Patents

Open source implementations are available, but the method is patented, which limits its free use.

MELP was developed by Texas Instruments (2400 bit / s, original version), Microsoft (transcoder of the 1200 bit / s version), Compandent (parts of the source code for all three bit rate modes ), Thales Group (600 bit / s version) and AT&T (Noise Reduction). Products and applications for the US government and NATO are exempt from any license claims. Use by third parties (including civil applications) is made almost impossible due to the lack of license management (comparable to MPEG LA ). Since the last change took place in 2005 (specification of the 600-bit / s version), patent protection is likely until 2025.

Web links