Copyright 1999 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE.

This article was published in the July 1999 issue of
IEEE Communications Magazine.

Abstract The introduction of Internet Protocol technology into traditional telecommunications networks is changing the nature of speech communication in those networks. As the current switched network infrastructure is augmented by packet networks, the needs of voice users of these hybrid networks must be given due consideration. These users are accustomed to high-quality connections with low distortion in the speech signal and low transmission delay in the speech path. In order for packetized networks to gain widespread acceptance for speech transmission services, it is necessary to maintain this high-quality performance in the evolving networks. In this article we discuss relevant speech transmission performance requirements and the associated activities taking place in regional and international standards bodies.

Speech Transmission Performance Planning in Hybrid IP/SCN Networks

Mark E. Perkins and Charles A. Dvorak, AT&T
Barry H. Lerich, Telcordia
Joseph A. Zebarth, Bell Canada

The evolution of speech telecommunications from the switched infrastructure of the current switched circuit network (SCN, including ISDN and cellular networks) to a packetized infrastructure is currently centered on Internet Protocol (IP) networks. As this evolution takes place, traditional voiceband services (i.e., those requiring a bandwidth of 300–3400 Hz) will be carried over hybrid IP/SCNs. In this article we review the impairments to speech transmission that occur in SCNs and those that occur in hybrid IP/SCNs. We give special attention to the existing standards literature on performance planning for speech transmission services. These standards are an accurate distillation of current knowledge and best practices in this area. The role of standards in ensuring interoperability of communications equipment is widely recognized. The role of standards in ensuring high quality speech transmission is often overlooked. Our intent is to show the communities of traditional telecommunications engineers and IP network engineers that answers to many common questions about speech transmission planning in hybrid networks are already available in the standards literature. To be sure, there are areas where there is still work to be done. To that end, we also take note of recent standards activities directed at ensuring that packet networks such as IP, as well as hybrid IP/SCNs, continue to provide the high-quality voiceband communications to which users of the current SCN have become accustomed.
We begin with a brief summary of factors affecting speech transmission quality, and give a brief description of existing and draft standards applicable to problems of speech transmission planning in hybrid networks. After summarizing hybrid IP/SCN architectures, we introduce the E-Model, a transmission planning tool that has shown great flexibility and utility when transmission planners must make recommendations about new network configurations. We next turn our attention to the kinds of impairments that can occur in the speech channel. Finally, we illustrate use of the E-Model to assess the effects of some of the impairments we have discussed.

Speech Transmission Quality in Telecommunications Networks

Transmission of speech has been the principal application in the SCN, and voice users of the SCN have come to expect high-quality transmission. The subjective performance (i.e., quality) of speech transmission is governed by a relatively small number of controllable parameters. The nature of the impairments introduced into a speech signal during transmission depends, to some extent, on the network over which the speech is carried. We first review the parameters affecting speech quality in the SCN, then focus on the speech quality issues faced by transmission planners of hybrid IP/SCNs. We close this section with a brief overview of recent standards activities relevant to planning speech services in hybrid IP/SCNs.

The SCN

Historically, the transmission quality of speech telecommunications has been dominated by three classes of impairments:

Loss (also known as loudness loss) -- reduction in signal strength that results in a received speech energy level that is too low
Noise -- circuit noise and other noise-like artifacts introduced by the transmission system
Echo -- either talker echo, where the talker has the experience of his/her voice returned after some transmission delay, or listener echo, where the listener has the experience of hearing an echo of the talker

The introduction of digital technology into the SCN has helped alleviate some of these impairments, but has also made possible the introduction of new classes of impairments. On one hand, loudness loss and circuit noise are minimized in digital networks. On the other hand, digital networks allow introduction of complex speech signal processing devices, resulting in either a net improvement of speech transmission quality (e.g., digital network echo cancellers can be used to reduce echo to levels virtually unnoticed by most users) or more efficient use of network resources (e.g., speech coding and other bandwidth compression techniques such as voice activity detection and silence removal). However, signal processing devices designed to improve network efficiency often have a negative impact on the subjective quality of speech transmissions (e.g., by increasing the absolute delay in the transmission path or by distorting the speech signal), especially when compared to that of the current SCN.

Hybrid IP/SCN Internetworks

In the future, the transmission quality of speech telecommunications services will be dominated by packetized networks. Voiceband services on packetized networks will provide the high quality performance expected by end users when service providers adopt 64 kb/s pulse code modulation (PCM) -- International Telecommunication Union -- Telecommunication Standardization Sector (ITU-T) Recommendation G.711 -- as the coding mechanism. In the near term, service providers are using low-bit-rate speech coding to make efficient use of available network bandwidth. Speech codecs such as ITU-T Recommendation G.729 provide high quality performance that may be acceptable for some applications. When low-bandwidth links are used widely to access these networks (e.g., access via a personal computer over a V.90 or V.34 modem link), the need for speech compression is essential. The impact of the combination of speech codec choice, packetization scheme, and other speech signal processing (e.g., voice activity detection, comfort noise insertion), when coupled with user expectations of high quality, will provide new challenges for telecommunication system design and implementation.
During the evolutionary phase from circuit-switched to packet-switched networks, internetworks composed of segments in each domain will be commonplace. While it is expected that data transmission will be the dominant source of traffic in these networks, voice communications will still be an important component of overall network traffic. It is therefore essential that the transmission performance needs of voice users of these networks be taken into account during all phases of network evolution. In particular, speech transmission must be accomplished over connections with low distortion and low delay.

Standards for Speech Transmission Performance in Hybrid Networks

New standards efforts are giving significant attention to network performance and planning issues affecting voice users of hybrid networks. At the regional level, T1A1.7 in North America and European Telecommunications Standards Institute (ETSI)/TIPHON in Europe have produced technical reports on this subject. These efforts have triggered international interest in ITU-T Study Group 12, which has the following suite of newly accepted Recommendations, draft Recommendations, and revised work projects to address these issues:

New Recommendation G.107 [1] describes the E-Model, a tool for assessing the relative impact of transmission planning decisions on speech quality.
Draft Recommendation G.108 (Application of the E-Model -- A Planning Guide) is a guide for transmission planning that includes various network scenarios (including hybrid networks), with examples of how to use the E-Model.
Draft Recommendation G.109 (Definition of Categories of Speech Transmission Quality) defines categories of speech quality in terms of output from the E-Model.
Draft Recommendation G.116 (Transmission Performance Objectives Applicable to End-to-End International Connections) provides general performance objectives for speech transmission in digital networks.
Draft Recommendation G.177 (Transmission Planning for Voiceband Services over Hybrid IP/SCN connections) offers specific guidance for addressing the myriad technical issues facing transmission planners of hybrid IP/SCNs.
The existing study topic on transmission performance for the interconnection of the SCN with other networks has been revised to take explicit account of hybrid IP/SCNs.
A new study topic on transmission performance effects of carrying traditional voiceband services on IP networks was approved in December 1998.

Hybrid IP/SCN Network Architectures

Industry convergence of systems design is using ITU-T Recommendation H.323 gateways to provide the needed interconnection functions between IP networks and the SCN. Figure 1 shows four possible IP/SCN connection arrangements. The terminals attached to the IP network are assumed to have H.323 functionality from the point of view of speech transmission. These terminals may be connected to the IP network via a direct connection (e.g., Ethernet, token ring) or a dial-up connection (e.g., modem and PPP link). The IP network and SCN sections are connected through the H.323 gateway. For convenience, this gateway is designated by a single box in Fig. 1. Decomposition of the gateway is a subject of discussion in ITU-T Study Group 16 and ETSI/TIPHON. Two new working groups in the Internet Engineering Task Force (IETF), Media Gateway Control (megaco) and Signaling Transport (sigtran), are defining protocols for use in the decomposed gateway. The details of this decomposition, some aspects of which are discussed in [2], are not the subject of this article. Rather, it is sufficient to note that there is a need for a signaling gateway (to transport signaling information between the two networks) and a media gateway (for protocol translation and transport of the voiceband signals between the two networks). In practice, the gateway may be composed of multiple pieces of equipment, each with specialized functions. The media gateway has significant impact on speech quality, since this is where much speech signal processing occurs.
The specific functions of the gateway will depend on whether the direction of transmission is from the IP network to the SCN or vice versa. In particular, the functions in the gateway include (but are not limited to):

IP network SCN
– IP packet disassembly
– Speech decoder (including error concealment, insertion of comfort noise or silence, etc.)
– Management or regulation of delay variation
– Echo cancellation
SCN IP network
– Speech encoder (including silence removal, comfort noise decisions, etc.)
– IP packet assembly
– Management or regulation of delay variation

Voiceband performance standards activities have addressed the four connection arrangements shown in Fig. 1. They are:

H.323 phone (H.323 IP network SCN phone)
Phone H.323 (phone SCN IP network H.323)
Phone phone (phone SCN * IP network SCN phone)
H.323 H.323 (H.323 IP network SCN IP network H.323)

Each of these arrangements requires at least one use of the gateway. Hence, connections that are purely SCN-based or purely IP-based have not been given significant attention in discussion of hybrid network performance. SCN-only connections have obviously been the subject of considerable attention for many years.
Of these four arrangements, types 1 and 3 are in use as service offerings today. Type 1 is a call initiated from an H.323 terminal to an SCN customer, as exemplified by various "PC-to-phone" services. Type 3 is a call from one SCN customer to another where, for example, the two local networks are circuit-switched and the long distance network uses IP. Types 1 and 2 use essentially the same facilities, but other aspects of the network are different for the two connection types. Type 2 connections require a means of translating between E.164 numbers (the standard telephone numbers used in the SCN) and IP addresses, a subject of study in the IETF and ETSI/TIPHON. Type 4 connections will occur as service providers migrate traditional telephony services to an IP infrastructure. One obvious example is a long distance call where the two local networks are IP-based, but the subscriber's long distance carrier uses a circuit-switched route.

Transmission Planning in Hybrid Networks Using the E-Model

The E-Model [1, 3] is a tool for assessing the relative impact that transmission planning decisions will have on speech performance. In general terms, the E-Model consists of additive terms that capture simultaneous impairments (e.g., environmental noise), delayed impairments (echo, long transmission times), and distortions attributable to specific pieces of transmission equipment (e.g., speech codecs). These general terms are composed from 18 parameters. The most important aspect of the E-Model is the introduction of an equipment impairment factor (eif). The eif has shown great utility in capturing the effects of new speech signal processing devices. In particular, the eif has been used to characterize the subjective impact of impairments such as IP packet loss [4]. Description of the eif methodology is found in [3, 5]. At the end of this article, we give the results of sample computations with the E-Model.
The output of the E-Model is a scalar factor, R, which is used to compare the speech transmission performance of alternative planning decisions. R can be mapped to estimates of customer opinion such as mean opinion score (MOS), percent good or better (%GoB), and percent poor or worse (%PoW). Use of R as a means for describing categories of speech quality, and how these categories relate to user satisfaction, is the subject of ITU-T Draft Recommendation G.109. Table 1 summarizes these mappings.

Speech Transmission Performance In Hybrid IP/SCNS

Speech quality in hybrid networks will be affected by the choice of speech coder, packet loss, long transmission delays (including any associated echo), and the implementation of other speech signal processing schemes (e.g., voice activity detection and comfort noise insertion). Since these are factors under the direct control of the network or service provider, we address them in turn.

Speech Coding

Modern speech codecs such as G.723.1 (5.3 or 6.3 kb/s), G.728 (16 kb/s), and G.729 (including Annex A, both at 8 kb/s) operate on segments of digitized speech known as frames and exploit predictable characteristics of speech to reduce the bandwidth usage in the transmission channel. The decision to use codecs such as G.729A or G.723.1 in hybrid IP/SCNs is based on a plan for efficient use of the available bandwidth (64 kb/s channel) and on the assumption that the user will want to be doing something else while talking (e.g., browsing or using IP-based videotelephony). If the aim is to provide conventional telephony services on an IP infrastructure, 64 kb/s PCM (G.711), 32 kb/s adaptive differential PCM (ADPCM) (G.726), or 16 kb/s low-delay code excited linear prediction (LD-CELP) (G.728) are better choices from many perspectives, most notably their lower delay, even in packetized applications. The critical point is that the choice of speech codec for current-generation services on hybrid networks must be made from a limited set, namely, G.729A and G.723.1. There are, of course, other codecs that operate at rates of 8 kb/s or less, but these are the only international standards in this range. We note that H.323 makes specific provisions for G.711, G.723.1, G.728, and G.729 (including Annexes A and B).
The use of speech codecs widely deployed in digital wireless systems (or newer codecs under consideration for such applications) should be considered for use in hybrid networks. While the transmission characteristics of wireless channels and those of IP networks may differ, speech coders for wireless applications are designed to be robust with respect to coded speech frames that have been corrupted and declared unusable. Hence, they exhibit desirable properties for speech coders deployed in hybrid networks. Moreover, since it can be expected that an increasing number of voice calls will include an IP/SCN gateway and a wireless terminal, use of the same codec by both systems is desirable. Whenever possible, only one encode/decode of speech should be performed, the objective being to reduce tandem processing by speech coders. To achieve this, it will be necessary to establish appropriate methods for communicating the type of codec in use so that the IP/SCN gateway does not perform unnecessary speech encoding or decoding in these situations.
The subjective performance of a speech codec is conveniently captured by the eif of the E-Model. The eif for a given codec is empirically (subjectively) determined and characterizes the subjective effect of speech impairments introduced by the codec. A given codec may have different eifs for operation under clear channel conditions and for different amounts of packet loss [4]. Hence, the eif provides a useful metric for comparing codecs on a scale of subjective performance (i.e., speech quality). Since eifs are additive, they also capture the effects of multiple codecs (whether of the same or different type) arranged in tandem. All other things being equal, a lower eif is better than a higher eif. For codecs with equal eifs, other considerations (such as delay or bit rate) may play decisive roles in codec choice. Current assignments of eifs for a wide range of speech codecs is available in [4].
The processing delay associated with a codec is also captured in the E-Model, but is not part of the eif assigned to the codec.

Effects of Transmission Errors and Packet Loss

In hybrid networks, two primary types of transmission errors may occur:

Bit errors on the transmission facility
Lost or discarded IP packets

Bit errors that occur on transmission facilities will ultimately result either in discarded packets (when errors occur in the IP header) or in corrupted payload (i.e., encoded speech). When errors occur in the encoded speech, codecs such as G.723.1, G.728, and G.729 include error concealment techniques that operate quite well, resulting in acceptable performance at bit error rates as high as 3 percent. Reliance on these techniques is preferable to retransmit requests, which will usually increase transmission delay.
The performance effect caused by lost or missing packets at the destination, and the associated effects on low-bit-rate speech coders, is of particular concern. To achieve even moderate efficiency of network utilization [2], it will be necessary to include multiple frames of coded speech in a single IP packet. This will, however, increase delay, so consideration must be given to the impact that packetization strategy has on delay. Thus, as noted above, speech codecs deployed in hybrid networks should be robust with respect to loss of multiple sequential frames. Speech codecs in the ITU-T G.700 series have been tested extensively and shown to have good quality and robust performance under a wide range of conditions. In particular, G.723.1, G.728, and G.729 have been evaluated under conditions that lead to random frame loss as well as "bursty" frame loss. However, they have not been evaluated thoroughly for one effect that may be peculiar to IP transmission, that is, the effect of loss of multiple sequential frames, either in isolation or in "bursty" situations. The impact of large percentages of frame loss (as much as 20–30 percent) also needs additional study.

Delay and Echo

Delay can have two effects on speech performance. First, it increases the subjective effect of any echo impairment. Second, as indicated in [6], even when echo is controlled, delays above 150 ms in each direction can interfere with the dynamics of speech conversation, depending on the type of conversation and degree of interaction required. Although the transmission delay of hybrid IP/SCNs may exceed the typical delay of the SCN, the benefits provided by new network and service capabilities might compensate for the degradation caused by the added delay. It is important to note that these trade-offs have not been quantified.
Current assessment of delay in hybrid IP/SCN connections indicates that echo control is required for all call scenarios shown in Fig. 1. ITU-T H.225.0 indicates that control of acoustic echo from an H.323 terminal is the responsibility of the terminal. In order to provide echo protection, all H.323-based terminals must provide acoustic isolation between the transmitter and the receiver. Sufficient isolation is achieved when the weighted terminal coupling loss [7] is at least 45 dB. Such acoustic isolation may be achieved relatively easily in standard handset terminals by careful design. In hands-free operation (e.g., microphone and speaker), more complex techniques are typically required.
As indicated in our architecture section, the gateway includes an echo canceller to cancel echoes from the SCN. At a minimum, these echo cancellers should meet the requirements in [8], including requirements that some voiceband data applications be allowed to disable echo cancellation. In some configurations, it is likely that such echo cancellers will work in tandem with SCN echo control devices. As described in [9], the gateway echo canceller should not degrade the overall echo control in the connection; nor should it add any significant degradation.

Other Bandwidth Reduction Techniques

Two common impairments in speech compression systems, not restricted to hybrid IP/SCNs, are temporal (syllable) clipping and noise contrast.
Temporal speech clipping is the loss of speech signal at any time, and can occur when, for example, voice activity detection is used. The subjective impact of clipping depends on four factors: duration of clip, percentage of speech clipped, frequency of clipping, and overall speech activity. Based on the results of detailed subjective tests[10], two guidelines (recently included in ITU-T Draft Recommendation G.116) to maintain good speech quality are:

Clipping of speech segments >= 64 ms should always be avoided.
Clipped segments < 64 ms should be kept below 0.2 percent of active speech.

When codecs such as G.729 and G.723.1 are used, packet loss will not necessarily result in speech clipping. When multiple sequential frames are lost, the result may be a period of unintelligibility rather than loss of speech energy. Which effect is preferable subjectively has not been studied.
Noise contrast occurs when background noise is interrupted due to digital speech processing, such as operation of voice activity detection (silence removal). Comfort noise is noise that is introduced to mask the negative effects of noise contrast.
For comfort noise insertion, the best subjective performance will be realized when the noise inserted at the receiving end matches, as closely as possible, the background noise at the sending end. Furthermore, the time course of changes in the level of the inserted noise should match, as closely as possible, the level changes that occur in the background noise. The voice activity detectors and comfort noise generators described in Annex B to Recommendation G.729 and Annex A to Recommendation G.723.1 both attempt to do this by extracting noise parameters at the sending end and transmitting them to the receiving end at a low bit rate, thus maintaining the bandwidth savings of silence removal.

Sample Computations Using the E-Model

In this section we show the results of sample computations using the E-Model to assess the impact of two important considerations in designing hybrid IP/SCN voice services:

The effect of packetization strategy
The combination of codec and transmission delay

Proper use of the E-Model requires knowledge of how a particular speech codec performs in the presence of packet loss with known statistics, packetization strategy, and any effect of buffering to remove variations in the time for packets to traverse the network. Limited information on eifs for these situations is available in [4]. Additional work is necessary.

Effect of Packetization Strategy

One-way delay in hybrid IP/SCNs is determined by the network delay, buffering delay, and packetization delay (which depends on the choice of speech codec and the number of encoded frames included in each IP packet).
Table 2 shows R-values for several scenarios. An "impairment-free" connection with 50 ms one-way delay is included as a reference. Such a connection meets the delay allocation defined in [6] for a national segment in an international connection. The only E-Model parameters that vary are codec and delay. All other parameters are set to the default values [1], meaning they have no effect on the computation. E-Model eifs for G.723.1 and G.729A are for the codec with voice activity detection turned on [4]. The "Packet delay" column is the time required to assemble a packet, and is computed as
T_F(N + 1) + T_L,
where T_F is the frame size for the codec, N is the number of frames in a packet, and T_L is the look-ahead time for the codec. For G.723.1 (both bit rates), the frame size is 30 ms and the look-ahead time 7.5 ms. For G.729A, the values are 10 ms and 5 ms, respectively. A duration of T_F + T_L (known as the algorithmic delay of the codec) is required to collect input for processing the first frame. We assume that the processor on which the codec is implemented is fully utilized (equivalently, that multiple encoder/
decoder pairs are implemented on the same processor). Hence, a single encoder/decoder pair requires T_F ms to encode and decode each frame in a packet. Note that we implicitly assume that the encoded packet is clocked instantaneously onto the transport facility. This is an optimistic, yet realistic, estimate of packetization delay. In practice, however, the actual delay should be no more than
T_F(2N + 1) + T_L.
The additional T_F ms (for each frame) will be required when the transport medium is operating at a rate equal to the output bit rate of the codec (e.g., 8 kb/s for G.729). If packetization delay is longer than this, it is because the encoder is not keeping pace with the input speech and loss of speech information will result. Sample computations are shown for up to 60 ms of speech per packet.

Effects of Codec Choice Combined with Transmission Delay

Table 3 shows R-values calculated for various combinations of speech codec and delay. Values of eif are taken from [4]. All other parameters of the E-Model were set to the default values [1]. In particular, no echo impairment is included. When packet loss is included, G.729A has two frames per packet and G.723.1 has one frame per packet. Packets were discarded at random.
Cells in the table are shaded corresponding to the R-value ranges shown in Table 1. Blackened cells indicate invalid combinations of delay and codec (i.e., the packetization delay of the codec alone is longer than the indicated one-way delay). R-values in the "G.711" column serve as convenient reference points for high-quality connections in the current switched network. None of the combinations of low-bit-rate codec and delay reaches the level of performance achieved in the SCN.

Summary and Conclusions

As hybrid IP/SCNs are deployed, the near-term use of low-bit-rate speech coding for speech transmission services will be more common than in the current SCN. Increased transmission delay, with the possibility of increased echo impairments in the speech channel, is also inevitable. New standards efforts are addressing the performance needs of voice users of evolving hybrid and packetized transmission systems. Transmission planners and service providers of speech services on hybrid networks will find significant utility in consulting the existing standards literature on performance planning for speech transmission in the SCN. Accelerated activities in ITU-T Study Group 12 have also addressed similar issues for hybrid networks, with final approval of new Recommendations expected later this year.
We have also identified areas that require additional study, in both the standards arena and the general area of acceptable levels of performance. Perhaps the most important area for additional work is to gain a better understanding of the effects of IP packet loss on speech codec performance and to have eifs for these conditions. The willingness of customers to tolerate longer transmission delays in exchange for enhanced services must be studied and quantified. Once sufficient data are available, the E-Model includes an "expectation factor" that may be used to adjust performance predictions to account for such tradeoffs. Other areas for study are in the realm of extensions and enhancements to the E-Model. The E-Model is grounded in extensive subjective testing. However, it is only applicable to traditional telephone-band communications between two humans, each of whom are using telephone handsets. Extensions to the E-Model are required to address hands-free operation as well as applications where automatic speech recognition and text-to-speech synthesis are used. The effects of speech codecs on certain speech-related applications (speaker recognition, automatic speech recognition, and text-to-speech) must be quantified and appropriate levels of performance must be set.

References
[1] ITU-T Rec. G.107, "The E-Model, a computational model for use in transmission planning," Dec. 1998.
[2] M. Hamdi et al., "Voice Service Interworking for PSTN and IP Networks," IEEE Commun. Mag., Special Issue on Interoperability of Networks for Interoperable Services, May 1999, pp. 104111.
[3] N. O. Johannesson, "The ETSI Computation Model: A Tool for Transmission Planning of Telephone Networks," IEEE Commun. Mag., Jan. 1997, pp. 7079.
[4] ITU-T Rec. G.113, App. I, "Provisional planning values for the equipment impairment factor," Dec. 1998.
[5] ITU-T Rec. G.113, "Transmission impairments," Feb. 1996.
[6] ITU-T Rec. G.114, "One-way transmission time," Feb. 1996.
[7] ANSI/TIA/EIA/579-A-98, "Telecommunications-Telephone Terminal Equipment-Transmission Requirements for Digital Wireline Telephones."
[8] ITU-T Rec. G.168, "Digital network echo cancellers," Apr. 1997.
[9] ITU-T Rec. G.131, "Control of talker echo," Aug. 1996.
[10] J. Gruber and L. Strawczynski, "Subjective effects of variable delay and speech loss in dynamically managed voice systems," Proc. IEEE GLOBECOM '82, vol. 2, Miami, FL, Nov.Dec. 1982, pp. F.7.3.15.

Biographies
Mark E. Perkins is a principal technical staff member at AT&T Laboratories where he manages standards activities for quality of service and architectural issues for IP-based services, and has been active in subjective performance evaluation of speech signal processing equipment. He serves as rapporteur in ITU-T Study Group 12 for a new project on speech transmission quality in IP networks. He earned a B.A. from the University of California, Riverside, and M.A. and Ph.D. degrees from New York University.
Charles A. Dvorak has been with AT&T since 1982. He heads the strategic standards division, which manages AT&T participation in telecommunications standards organizations. He has also supervised the Voice Quality Assessment Lab of Bell Laboratories. He was the chair of standards body T1A1 for four years, and is currently the vice-chairman of ITU-T Study Group 12. He received M.E.E. and Ph.D. degrees from the University of Delaware in 1975 and 1978.
Barry H. Lerich is a project director in the data communications and transport integration organization of Telcordia Technologies (formerly Bellcore) where he develops performance requirements for voice and data communications services. He has been active in telecommunications standards activities of T1A1.7, Signal Processing and Network Performance for Voice and Voiceband Data, and T1M1.3, OAM&P Testing. He currently serves as chair of Working Group T1A1.7. He earned a B.S. from Purdue University and an M.S. from the Naval Postgraduate School, Monterey, California.
Joseph A. Zebarth is an associate director at Bell Canada. He earned a BA Sc. from the University of Waterloo. He has held positions in standards development, transport system design, network planning, outside plant engineering, and equipment provisioning. Currently he is responsible for the development of strategic internet and multimedia transmission performance standards. He is rapporteur in ITU-T Study Group 12 for projects on the introduction of ATM technology into networks and on the impact of delay and echo on Internet and multimedia services.