Clifford J. Weinstein

Experience with Speech Communication in Packet Networks

The integration of digital voice with data in a common packet-switched network system offers a number of potential benefits, including reduced systems cost through sharing of switching and transmission resources, flexible internetworking among systems utilizing different transmission media, and enhanced services for users requiring access to both voice and data communications. Issues which it has been necessary to address in order to realize these benefits include reconstitution of speech from packets arriving at nonuniform intervals, maximization of packet speech multiplexing efficiency, and determination of the implementation requirements for terminals and switching in a large-scale packet voice/data system. A series of packet speech systems experiments to address these issues has been conducted under the sponsorship of the Defense Advanced Research Projects Agency (DARPA). In the initial experiments on the ARPANET, the basic feasibility of speech communication on a store-and-forward packet network was demonstrated. Techniques were developed for reconstitution of speech from packets, and protocols were developed for call setup and for speech transport. Later speech experiments utilizing the Atlantic packet satellite network (SATNET) led to the development of techniques for efficient voice conferencing in a broadcast environment, and for internetting speech between a store-and-forward net (ARPANEI) and a broadcast net (SATNET). Large-scale packet speech multiplexing experiments could not be carried out on ARPANET or SATNET where the network link capacities severely restrict the number of speech users that can be accommodated. However, experiments are currently being carried out using a wide-band satellite-based packet system designed to accommodate a sufficient number of simultaneous users to support realistic experiments in efficient statistical multiplexing. Key developments to date associated with the wide-band experiments have been 1) techniques for internetting via voice/data gateways from a variety of local access networks (packet cable, packet radio, and circuit-switched) to a long-haul broadcast satellite network and 2) compact implementations of packet voice terminals with full protocol and voice capabilities. Basic concepts and issues associated with packet speech systems are described. Requirements and techniques for speech processing, voice protocols, packetization and reconstitution, conferencing, and multiplexing are discussed in the context of a generic packet speech system configuration. Specific experimental configurations and key packet speech results on the ARPANET, SATNET, and wide-band system are reviewed.

Fractional Speech Loss and Talker Activity Model for TASI and for Packet-Switched Speech

The cutout fraction in a TASI system is shown to be \phi=\frac{1}{np}\sum_{k=c+1}^n (k-c) n\choose k p^{k}(1-p)^{n-k} where n is the number of sources, c is the number of channels, and p is the probability that a source is issuing a talkspurt at a random time. This result is shown to hold independently of the probability density function of talkspurt duration. The same formula is shown to apply to the fraction of packets lost in a packet-switched link with a transmission capacity of c packets every T p seconds, where T p is the interval between packet generations for an individual source during talkspurt, and where no packet is queued for a time longer than T p . In addition, a simple Markov birth-death model is presented for the random process a(t) representing the number of talkers issuing talkspurts at a given time.

Data Traffic Performance of Integrated Circuit- and Packet-Switched Multiplex Structure

Results are developed for data traffic performance in an integrated multiplex structure which includes circuit-switching for voice and packet-switching for data. The results are obtained both through simulation and analysis, and show that excessive data queues and delays will build up under heavy loading conditions. These large data delays occur during periods of time when the voice traffic load through the multiplexer exceeds its statistical average. A variety of flow control mechanisms to reduce data packet delays are investigated. These mechanisms include control of voice bit rate, limitation of the data buffer, and combinations of voice rate and data buffer control. Simulations indicate that these flow control mechanisms provide substantial improvements in system performance.

Roundoff noise in floating point fast Fourier transform computation

A statistical model for roundoff errors is used to predict output noise-to-signal ratio when a fast Fourier transform is computed using floating point arithmetic. The result, derived for the case of white input signal, is that the ratio of mean-squared output noise to mean-squared output signal varies essentially as \nu = \log_{2}N where N is the number of points transformed. This predicted result is significantly lower than bounds previously derived on mean-squared output noise-to-signal ratio, which are proportional to ν2. The predictions are verified experimentally, with excellent agreement. The model applies to rounded arithmetic, and it is found experimentally that if one truncates, rather than rounds, the results of floating point additions and multiplications, the output noise increases significantly (for a given ν). Also, for truncation, a greater than linear increase with ν of the output noise-to-signal ratio is observed; the empirical results seem to be proportional to ν2, rather than to ν.

Exploiting Nonacoustic Sensors for Speech Encoding

The intelligibility of speech transmitted through low-rate coders is severely degraded when high levels of acoustic noise are present in the acoustic environment. Recent advances in nonacoustic sensors, including microwave radar, skin vibration, and bone conduction sensors, provide the exciting possibility of both glottal excitation and, more generally, vocal tract measurements that are relatively immune to acoustic disturbances and can supplement the acoustic speech waveform. We are currently investigating methods of combining the output of these sensors for use in low-rate encoding according to their capability in representing specific speech characteristics in different frequency bands. Nonacoustic sensors have the ability to reveal certain speech attributes lost in the noisy acoustic signal; for example, low-energy consonant voice bars, nasality, and glottalized excitation. By fusing nonacoustic low-frequency and pitch content with acoustic-microphone content, we have achieved significant intelligibility performance gains using the DRT across a variety of environments over the government standard 2400-bps MELPe coder. By fusing quantized high-band 4-to-8-kHz speech, requiring only an additional 116 bps, we obtain further DRT performance gains by exploiting the ears insensitivity to fine spectral detail in this frequency region.

The Tradeoff Between Delay and TASI Advantage in a Packetized Speech Multiplexer

A packetized speech multiplexer differs from a circuitswitched TASI system in that the presence of a packet buffer allows a tradeoff where the TASI advantage can be increased at a cost in packet delay. This tradeoff is investigated via a simulation. Results are presented to show the relations between TASI advantage and delay, for both an average delay criterion and a maximum delay criterion. It is shown that, particularly for the case where small numbers of talkers are multiplexed, the packetized system offers significant improvements in TASI advantage over the conventional circuit-switched multiplexer, at modest costs in packet delay.

Measuring human readability of machine generated text: three case studies in speech recognition and machine translation

We present highlights from three experiments that test the readability of current state-of-the art system output from: (1) an automated English speech-to-text (SST) system; (2) a text-based Arabic-to-English machine translation (MT) system; and (3) an audio-based Arabic-to-English MT process. We measure readability in terms of reaction time and passage comprehension in each case, applying standard psycholinguistic testing procedures and a modified version of the standard defense language proficiency test for Arabic called the DLPT*. We learned that: (1) subjects are slowed down by about 25% when reading system STT output; (2) text-based MT systems enable an English speaker to pass Arabic Level 2 on the DLPT*; and (3) audio-based MT systems do not enable English speakers to pass Arabic Level 2. We intend for these generic measures of readability to predict performance of more application-specific tasks.

The Experimental Integrated Switched Network - a System-Level Network Test Facility

An Experimental Integrated Switched Network (EISN) has been developed to provide a system-level testbed for the evaluation of advanced communications networking techniques, including survivable network routing algorithms using a mix of transmission media, for application in the Defense Switched Network (DSN). EISN includes five CONUS sites linked by a wideband demand-assigned satellite channel and by dialed-up terrestrial trunks for alternate satellite/terrestrial routing experiments. Experiments to date have validated techniques for integration of circuit-switched terrestrial systems with the demand-assigned satellite system, and for the establishment of alternate routes over satellite and terrestrial paths. Currently, candidate routing algorithms for application in the DSN are being implemented and tested using external routing/controller processors attached to digital circuit switches at EISN sites. In addition, EISN is also being used to support data communication experiments using DoD standard data protocols in a combined satellite/ terrestrial network environment. Work is ongoing both in system experiments and in testbed developments to include additional capabilities. This paper represents a description and status report on both the testbed and the experimental efforts.

Predictive coding in a homomorphic vocoder

Application of a type of predictive coding to the channel signals of a homomorphic vocoder has produced sizable bit rate reductions. With only slight degradation in speech quality, reduction (for the spectral envelope information) from 7800 to 4000 bits/s was achieved. A technique for obtaining the formant frequencies from the predictive coding parameters is described; this approach promises further bit rate reductions. As a by-product of this study of predictive coding, direct and cascade form speech synthesizers are compared on the basis of differing quantization effects.

Multisensor MELPe using parameter substitution

The estimation of speech parameters and the intelligibility of speech transmitted through low-rate coders, such as MELP (mixed excitation linear prediction), are severely degraded when there are high levels of acoustic noise in the speaking environment. The application of nonacoustic and nontraditional sensors, which are less sensitive to acoustic noise than the standard microphone, is being investigated as a means to address this problem. Sensors being investigated include the general electromagnetic motion sensor (GEMS) and the physiological microphone (P-mic). As an initial effort in this direction, a multisensor MELPe coder (MELP coder with the addition of a noise preprocessor) using parameter substitution has been developed, where pitch and voicing parameters are obtained from GEMS and P-Mic sensors, respectively, and the remaining parameters are obtained as usual from a standard acoustic microphone. This parameter substitution technique is shown to produce significant and promising DRT (diagnostic rhyme test) intelligibility improvements over the standard 2400 bps MELPe coder in several high-noise military environments. Further work is in progress aimed at utilizing the nontraditional sensors for additional intelligibility improvements and for more effective lower-rate coding in noise.