Return to DDVPC HomePage Return to TEC homepage
 
Download Entire Document in Postscript Format (126K)

Audio Examples of Tandem and Single Coder Processing in 8kHz WAV format:

Effects of Tandeming on Voice Quality and Intelligibility

Tandems occur whenever the output of one type of voice coder is used as input for a second coder. This happens when people using different types of coders need to talk to each other, for example, when a person using an ANDVT terminal running LPC-10 at 2.4 kbps wants to talk to someone with a VINSON KY-57 running CVSD at 16 kbps, or, in the commercial world, when a Bell Atlantic cell phone customer wants to talk to someone with a Sprint Spectrum cell phone. At reasonably high data rates, although voice quality does suffer somewhat, the consequences are not too awful. But tandems with low data rate voice coders lead to significant losses in intelligibility and voice quality.

Tandems may be of two types. The digital speech from one coder may be converted back to analog before being redigitized and analyzed by the second coder. Or the digital signal from one coder may be decoded but remain in digital form, and the digital waveform is then used as the input to the next coder. Staying in the digital domain generally leads to less loss in voice quality than going back to analog speech. Except for very high quality digital coders, there will be some degradation from one coder to the next, but the result is considerably less degraded with a digital interface than with an analog interface. Many gateways today have only analog interfaces. Digital tandems, where the input coder is of high quality (e.g., 64 kbps PCM), may cause little or no additional loss in voice quality. However, if the first coder degrades the speech, then this degradation will cause additional losses for the second coder, especially if the second coder is a low data rate coder, such as LPC at 2.4 kbps.

If the digital output of a speech coder is sent as data (like bits over a computer modem) and the bit stream is maintained intact until it is finally decoded at the receiver by the same type of coder, then sending the bit stream over several different channels will not degrade the speech unless channel errors are introduced. Sending LPC coded speech over several links (e.g., a phone line and a satellite link) to a receiver that decodes the LPC bit stream will not of itself reduce voice quality (no tandem has occurred).

A tandem can never be better than its weakest link and is often worse. In many cases, especially if there are multiple tandems, it will be considerably worse. The use of multiple tandems can quickly lead to voice links that are for practical purposes unintelligible. When the voice quality sinks below minimum acceptable levels, users tend to respond by using nonsecure communication links. If you can't make yourself understood, the system is of no use.

The table to the right lists Diagnostic Rhyme Test (DRT) scores for tandems of CVSD (Vinson) with existing and new standard voice coders. These scores were obtained in the selection tests that were conducted by the DoD Digital Voice Processing Consortium (DDVPC) in 1994 to 1996, that resulted in the selection of MELP as the new 2.4 kbps standard[1]. At that time, tandems were tested only in the direction that was known to cause the most serious problems. The table at the end of this note can be used to help interpret these scores. Tandems in these tests were digital, except that tandems into CVSD were analog.

These scores are for "ideal" conditions, i.e., no background noise and error free channel conditions. Any additional degradations cause the scores to be reduced even more, and the problems are rapidly compounded when multiple tandems are used.

The following recommendation was made in the conclusions and in the executive summary of the 1984 DDVPC Final Report[2]. This recommendation is even more relevant in today's communication environments, where the use of mixed commercial channels is likely to result in multiple tandeming.

"Tandem operation of voice processors be held to an absolute minimum, including gateways and conference bridges. Gateways between systems should be implemented with the DoD standard LPC-10 using the 2.4 kbps signal; all voice terminals in higher bit-rate systems should also contain a DoD Standard LPC-10 voice processor. Analog conference bridges should be avoided.

"Tandem operation should probably be avoided altogether when acoustic background noise is present or high bit error rates are expected."

The results in the table above are consistent with the 1984 DDVPC tandem test results (all tandems at that time were analog). The 1984 tests show clearly that going from a high data rate to a lower one degrades the voice much more than going from the low data rate to the higher one. In either case, the tandem is significantly degraded and is worse than the poorer of the two systems by itself. Intelligibility for LPC-10 dropped from a DRT score of 88.2 under ideal conditions to 75.1 for 16 kbps CVSD into LPC and to 77.8 for LPC into CVSD at 16 kbps. Quality, as measured by the Diagnostic Acceptability Measure (DAM), showed similar losses. Quality for LPC-10 dropped from a DAM score of 46.9 under ideal conditions to 34.8 for 16 kbps CVSD into LPC and to 41.2 for LPC into CVSD at 16 kbps.

Tandems with existing military coders, such as CVSD, are almost certain to have reduced voice quality. Different commercial vendors use different speech algorithms. Some modern algorithms at 16 kbps or less may be of sufficiently high quality to cause little or no degradation in a digital tandem, but unless intelligibility and quality tests are conducted and show no reduction in scores, there is no guarantee that the voice quality will be acceptable when tandems occur.

The following table gives rough equivalence categories for describing DRT and DAM scores in everyday language.

References

[1] Bielefeld, M., Champion, C., Gatewood, P., Kohler, M. A., Kreamer, E. W., LaFollette, P., Scmidt-Nielsen, A., Supplee, L., and Tardelli, J., (1997). Selection of the 1996 Federal Standard 2400bps Voice Coding Algorithm. R2 Technical Report no. R22-008-97. National Security Agency, Ft. Meade, MD.
[2] Sandy, G. F., & Parker. (1984). Digital Voice Processor Consortium Final Report MTR-84W00053. McLean, VA, MITRE Corp.

 This website is sponsored by ARCON Corporation for the DDVPC and as such must conform to DoD policies concerning websites. Please read this disclaimer for more information.

Comments concerning design and content of these pages  should be sent to info@arcon.com. This page was last updated on 26-Oct-2001.