A New Federal Standard Algorithm for 2400bps Coded Voice
The DOD Digital Voice Processor Consortium (DDVPC) announced the Mixed Excitation Linear Prediction (MELP) algorithm as the new speech encoding Federal Standard at 2400 bps. This standard will replace the LPC-10 (Linear Predictive Coding) technology developed in the early 1970's. The DDVPC selected Rome Laboratory's Electromagnetics and Reliability Directorate, ERC-1 Office at Hanscom AFB, MA as the Host Laboratory for this effort.
The candidate speech coders underwent both standard evaluation, for voice quality and speech intelligibility (Mean Opinion Score and Diagnostic Rhyme Test), as well as newly developed tests designed to evaluate criteria of keen interest to the DOD, but which the DDVPC felt would not be adequately tested by the standard techniques. These include system usability under operational conditions, for which Rome Laboratory developed a Communicability test, and Talker Recognizability, a test developed by the Naval Research Laboratory.
The following papers were presented in special session at the 1996 IEEE International Conference on Acoustics, Speech & Signal Processing (ICASSP-96) , May 7 -10, Atlanta, GA, USA:
- Philosophy and Goals of the DoD 2400 bps Vocoder Selection Process
- Developing a Test Program for the DoD 2400 bps Vocoder Selection
- Vocoder Intelligibility and Quality Test Methods
- Speaker Recognizability Testing for Voice Coders[external link]
- Communicability Testing for Voice Coders
- Host Laboratory Functions for the DoD 2400bps Vocoder Selection
- Criteria for the DoD 2400bps Vocoder Selection
Results of the performance and complexity evaluations and the selection of a DDVPC recommended coder for the new Federal Standard were presented during a Panel Discussion at ICASSP-96. The results and decisions are:
- Coder Requirements Results
- Phase I Intelligibility Results
- Phase I Quality Results
- Phase I Overall Performance Results
- Combined Intelligibility Results
- Combined Quality Results
- Speaker Recognizability Results
- Communicability Results
- Overall Performance Results
- Coder Complexity Calculations
- Coder Figure of Merit & DDVPC Selection
Download all 11 documents in PostScript format (ddvpcres.zip - 74K)
The results for the final four coders including male and female talker averages and standard error values are presented in the following tables:
- Overall Intelligibility Performance
- Overall Quality Performance
- Speaker Recognizability Performance
- Communicability Performance
- Overall Performance
Download all tables in PostScript format (final4.zip - 32K)
Sample speech from the Phase II candidate coders, CELP, and unprocessed for the Quiet, Office, E-3-A airceaft and HMMWV environments is available for listening in the following table. The files are in (.wav) format and are 8K sampled, 16 bit linear PCM representations of actual Harvard sentence pairs used for the coder evaluations. Male (M) and Female (F) talker samples are available for each case.
| Null - unprocessed original | |||
| Quiet/Dynamic Mic | M | F | |
| Office/STU-III Handset | M | F | |
| E-3-A/R215 Headset | M | F | |
| HMMWV/H250 Handset | M | F | |
| FS1016 Celp (2400 bps) | |||
| Quiet/Dynamic Mic | M | F | |
| Office/STU-III Handset | M | F | |
| E-3-A/R215 Headset | M | F | |
| HMMWV/H250 Handset | M | F | |
| MELP (2400 bps) | |||
| Quiet/Dynamic Mic | M | F | |
| Office/STU-III Handset | M | F | |
| E-3-A/R215 Headset | M | F | |
| HMMWV/H250 Handset | M | F | |
| DVSI (2400 bps) | |||
| Quiet/Dynamic Mic | M | F | |
| Office/STU-III Handset | M | F | |
| E-3-A/R215 Headset | M | F | |
| HMMWV/H250 Handset | M | F | |
| ATT (2400 bps) | |||
| Quiet/Dynamic Mic | M | F | |
| Office/STU-III Handset | M | F | |
| E-3-A/R215 Headset | M | F | |
| HMMWV/H250 Handset | M | F | |
| MIT/LL (2400 bps) | |||
| Quiet/Dynamic Mic | M | F | |
| Office/STU-III Handset | M | F | |
| E-3-A/R215 Headset | M | F | |
| HMMWV/H250 Handset | M | F |
Additional details on the ICASSP-96 presentations can be found on the ICASSP-96 CD-ROM.
