Hi everyone, I read that MELPe won the DoD & NATO competitions against many vocoders such as AMBE, EMBE, STC, FSHC, HSX, SB-LPC, CVSD, CELP and LPC-10e, so it has to be clearly better than all of them. See for example at: http://www.compandent.com/melpe_faq.htm#Q06 Although MOS is a very comprehensive and expensive subjective test, it seems like marketing people may have been "shaping" the MOS of their vocoders or may have confused MOS with some software measure such as PESQ, and advertised them as having MOS scores that people find as hard to believe in... since if you would have taken their scores seriously they shouldn't lose the DoD & NATO competitions against MELPe afterall... BTW, does anyone know what is the is the MOS (Mean Opinion Score) of MELPe? (I mean a true MOS test! not PESQ or any other "score") Thanks, JM
MELPe MOS vs other vocoders
Started by ●March 28, 2008
Reply by ●April 8, 20082008-04-08
Indeed MELPe is better than AMBE, EMBE, STC, FSHC, HSX, SB-LPC, CVSD, CELP and LPC-10e! But some marketing people, who know very little or nothing about MOS would sometimes "shape" their vocoder's MOS or may confuse MOS with some software measure such as PESQ, and advertised them as having MOS scores that people find as hard to believe in... When you test the PESQ of vocoder that has noise-preprocessing (such as MELPe has), eventhough the noise-prepeocessing would definitely improve the quality of noisy coded speech and hence the MOS, the PESQ on the other hand would mistakenly indicate "poor matching" or "low quality" since the noise was removed.... :) So PESQ is good for those who want to believe in it... It can take speech coding researcher months of hard work (or even years) to get improvement of a fraction of a MOS score, but it can take only a few seconds for a marketing guy to get that... :) So if someone tells you that any of the above vocoders is better than MELPe, I would recommend you to not waste another second on that guy... Compandent has taken the MELPe to the next level, and their MELPe++ suite seems to sound great for bit-rates of 2400 / 1200 / 600 bps . It is great for secure voice, secure radio, satellite phones, wireless phones, tough condition radio device, long-range cordless phones, answering devices, VoIP, dictionaries & translators, and long-time players. Also, see: http://www.compandent.com/products_melpe.htm MELPe - Enhanced Mixed-Excitation Linear Predictive Vocoder Compandent's MELPe (MELPe-Plus-Plus) produces better quality than the standard reference! Compandent has been supporting and improving the MELPe real-time implementation, and has found and fixed numerous bugs in it, many of these bugs could not be discovered by simply processing the standard Test Vectors, rather than by comprehensive process of analyzing and optimizing the code and testing it under wide variety of scenarios and conditions conducted by Compandent's speech coding and DSP experts. We can show you examples where the standard reference produces hoarse speech or annoying clicks while Compandent's MELPe (MELPe++) produces much better sounding and cleaner speech. Therefore Compandent's MELPe (MELPe++) is in fact better than the standard reference! Compandent has also greatly improved and optimized the code. Compandent has ported MELPe to various DSPs by Texas Instruments such as TMS320c54xx, TMS320c55xx, TMS320c64xx, and OMAP. The MELPe-Plus-Plus software suite includes the following features: hand optimized Assembly real-time implementation of all algorithm components extensively analyzed, tested, fixed, and improved real-time implementation C-callable high-level functions optional components available (customer can select desired combination): MELPe at 2400 bps - high rate encoder and/or decoder MELPe at 1200 bps - low rate encoder and/or decoder MELPe at 600 bps - very low rate encoder and/or decoder compressed bit-stream transcoding between the rates noise pre-processor - for reducing background noise postfilter - for quality enhancement of the reproduced speech multi-channel implementation example of main C program that initializes and runs the MELPe vocoder functions "Packetized-Network-Ready" - to maintain high quality even in severe FER conditions operation mode can be switched on the fly (no need to reload the program to the DSP) easy and fast Test Vectors running and verification setup comprehensive and spectacular Real-Time demo, using DIP-switch control and LEDs indicators comprehensive and detailed documentation that allows for smooth and easy integration Compandent support and service very low-cost (about $400) complete DSP development board and tools (CCS) on which Compandent's MELPe may run, process the Test Vectors, and spectacular real-time audio demo System solution FLEXI-232 and FLEXI-PC104 : Cost-effective board and software for customized communication link to your modem or other equipment. John Mammoth wrote:> Hi everyone, > > I read that MELPe won the DoD & NATO competitions against many vocoders > such as AMBE, EMBE, STC, FSHC, HSX, SB-LPC, CVSD, CELP and LPC-10e, so > it has to be clearly better than all of them. See for example at: > http://www.compandent.com/melpe_faq.htm#Q06 > > Although MOS is a very comprehensive and expensive subjective test, it > seems like marketing people may have been "shaping" the MOS of their > vocoders or may have confused MOS with some software measure such as > PESQ, and advertised them as having MOS scores that people find as hard > to believe in... since if you would have taken their scores seriously > they shouldn't lose the DoD & NATO competitions against MELPe afterall... > > BTW, does anyone know what is the is the MOS (Mean Opinion Score) of > MELPe? (I mean a true MOS test! not PESQ or any other "score") > > Thanks, > JM
Reply by ●August 28, 20082008-08-28