Analog Speech Research Articles

Analysis of buffer behaviour for an integrated voice-data system

The letter considers an integrated voice-data system in which data are synchronously multiplexed in analogue speech. The behaviour of the data buffer of this system is studied in a more realistic way than in previous analyses, which leads to considerably different results.

Analog Scramblers for Speech Based on Sequential Permutations in Time and Frequency

Permutation of speech segments is frequently utilized in scramblers for analog speech privacy. This paper discusses a “sequential” permutation procedure that has better segment-separation properties than the well-known procedure of “block” permutation, where contiguous segments are arranged in blocks of appropriate size, and permuted within such blocks. It further proposes the application of the sequential procedure to a novel technique for simultaneous permutations in time and frequency. The paper also presents results of a subjective experiment where we measured residual speech intelligibility at the output of scramblers using permutations in time [time segment permutation (TSP)], or permutations in time and frequency [time-frequency segment permutation (TFSP)]. The experiment included examples of block TSP, sequential TSP, and sequential TFSP. We measured spoken-digit-intelligibility as a function of the communication delay introduced by the scrambling operation. We found that even with a delay of 512 ms, the residual intelligibility in a TSP scrambler is no lower than about 50 percent; however, a sequential TFSP scrambler can realize an average digit intelligibility in the order of 20 percent with a delay of 256 ms. A companion paper discusses the implementation of the sequential TFSP scrambler, and the quality of descrambled speech in the context of real-channel operation.

Intelligibility comparisons for two synthetic and one natural speech source

A set of simple test procedures for making a quantitative assessment of synthetic speech intelligibility characteristics at the phonentic level is described. To show where the major weaknesses of the synthesized speech sources lie, comparisons were made between natural speech, analysis-synthesis speech, and speech synthesized by rule, using the same test materials and protocol. The results indicate that the intelligibility of synthesized speech sources is equal or superior to natural speech for vocalic segments, but that synthetic stop and fricative class consonants have lower intelligibility than their natural counterparts. Moreover, the synthesized consonants were more vulnerable to degradation by masking noise, showing the synthesized acoustic cures to be less robust perceptually. The results suggest that terminal analogue speech synthesizers perform least well when generating approximations to the complex spectra found in fricatives and the release phase of stops. Not unexpectedly, the rule-synthesized consonants showed poorer performance in noise than the analysis-synthesis consonants, since the derivation of their parametric specifications was one step further removed from the natural speech parametric data.

Analog scramblers for speech privacy

Despite significant strides in digital speech technology, scrambling of the analog speech signal continues to be an important method for achieving casual privacy in many classes of voice communications systems. This paper explains the principles of important classes of analog speech scramblers, with emphasis on algorithms that possess demonstrated robustness in the context of real-channel operation. At the forefront of the art are so-called two-dimensional analog scramblers that manipulate the speech signal in both time and frequency domains, to obtain impressive combinations of residual intelligibility, communication delay and transmission robustness. Our review includes several examples of such two-dimensional algorithms.

Articulatory parameters for the perception of bilabials.

In order to investigate articulatory cues for the perception of bilabial consonants, a production model was set up whose control parameters were converted into acoustic parameters by which a terminal analogue speech synthesizer was driven to synthesize test sounds for the perception experiments. Experiments concerning discrimination between (b) and (w) and between (m) and (w) were then performed with these test sounds. Statistical testing of the results showed that discrimination among the bilabials in an intervocalic context entirely depended on the minimum value of lip opening and that, contrary to the facts hitherto known, it was independent of the rate of lip opening/closure. At the same time, the features [+ anterior, - coronal] and [+ nasal] turned out to be perceptually independent phonological attributes of bilabial and nasal consonants, respectively.

The elicitation and habituation of orienting and defensive responses to phobic imagery and the incremental stimulus intensity effect.

ABSTRACTUsing imaginal stimuli, the present investigation examined psychophysiological responses to imagery and the roles of Groves and Thompson's (1970) dual process theory and Sokolov's (1963) model of the defensive response(DR) in mediating the incremental stimulus intensity effect (ISIE). Experiment I was undertaken to determine whether phobic and neutral imagery could be discriminated by DRs and orienting responses (ORs). In a within groups design. 19 analogue speech phobics were exposed to 4 neutral and 4 phobic scenes. Dependent measures were skin conductance measures and second‐by‐second heart rate (HR). The results showed that DRs occurred only in response to phobic scenes. In a between groups design. Experiment II compared an incremental series (IS) of 16 phobic imaginal stimulus presentations to a constant series. Subjects were analogue speech phobics with 14 subjects per condition. For both conditions the last 4 scene presentations were the same. Dependent measures were second‐by‐second HR, second‐by‐second cephalic pulse amplitude, and skin conductance measures. The results showed greater habituation in the IS condition, but were inconclusive with regard to the roles played by dual process theory and the DR model in the ISIE.

A survey of methods for digitally encoding speech signals

Modern communications systems are increasingly using digital transmission, at various digit rates, for the many advantages thereby obtained. there are many methods that can be used for converting analogue speech signals to digital form. the aim is to produce very good speech quality using simple equipment at a low digit rate, but any application inevitably involves choosing the appropriate compromise between these three conflicting attributes. the paper reviews the most important of currently known methods, including systems already in widespread use and those still in the research stage. thel problem of assessing speech transmission performance is also discussed.

Buffer analysis for asynchronous data interpolation in analogue speech

The store-and-forward buffer required for asynchronous data interpolation (ADI) in analogue speech is analysed using a queueing model with single server having 1st-order Markov interruptions. The buffer-content probability generating function and probability mass function are derived. It is shown that the empty-buffer probability in the finite buffer equals that of the infinite buffer with its offered load equal to the carried load of the finite buffer. This concept proves helpful in obtaining a solution for the finite buffer through the solution for infinite buffer which is obtained in a closed form. This results in considerable saving in computer time. Although the results presented are for analogue speech, they can serve as a worst-case design for the STDM case.

Single server queueing model for data multiplexing in analog speech

Finite buffer behavior of an asynchronous time division multiplexed (ATDM) channel is investigated. Mixed data comprising of single character and burst (String of characters) is asynchronously multiplexed in analog speech on the channel. A queueing model with limited buffer capacity, mixed input traffic (Poisson and burst Poisson arrivals), constant transmission rate (while serving) and server interruptions through a first order Markov process is studied for the purpose. Our results indicate that for a given traffic intensity, buffer behavior in this system lies between that for Poisson input arrivals and that for burst Poisson input arrivals. Therefore, the buffer should be designed for the pure burst Poisson arrivals as it may be treated as the worst case design for the pure Poisson and the mixed arrivals as well.

A manpack satellite communications earth station

Recent advances in technology have enabled a family ofsmall microwave satellite communication terminals to be constructed. This paper describes one of these, a prototype manpack equipment which can be carried by asingle person. It was completed in mid-1978 and has since been demonstrated on numerous occasions.Using a suitable satellite, such equipment can be usedto establish a link with a main SATCOM base station; alternatively manpack/manpack links are possible provided here is a base station rebroadcast. Traffic facilities are either 50 baud duplex telegraph or analogue speech.

A Comparison of Four Methods for Analog Speech Privacy

Four well-known procedures for analog speech privacy have been compared in terms of residual intelligibility, bandwidth expansion, and encoding delay. Intelligibility scores have been determined from a perceptual experiment where about 70 untrained listeners were given the task of recognizing each of 200 spoken digits that occurred in a balanced set of 50 encrypted four-digit utterances, and by averaging resulting probabilities of correct digit recognition. Bandwidth expansion has been expressed in terms of a new segmental measure that is more sensitive to short-time bandwidth manipulations than a conventional, long-time-averaged power spectrum measurement. Encoding delay is a straightforward function of analog scrambler parameters. The scrambling procedures that have been compared are sample permutation ( <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">S</tex> ), block permutation ( <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">B</tex> ), frequency inversion ( <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">F</tex> ), and a combination of methods <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">B</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">F</tex> , denoted by [ <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">BF</tex> ]. Sample permutations involved a contiguous set of L <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">S</inf> (2 to 128) 8 kHz samples, while block permutations operated on a contiguous set of N <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">B</inf> (4 to 128) speech segments each of which was L <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">B</inf> (8 to 256) samples long. Frequency inversion is obtained by simply inverting the sign of every other Nyquist (8 kHz) sample. The parameters, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">L_{s},N_{B}</tex> , and L <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">B</inf> , determine residual intelligibility as well as transmission properties such as encoding delay and bandwidth. The comparisons in our study provide a quantitative justification for the popular approach [ <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">BF</tex> ]. For example, with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">N_{B} = 8</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">L_{B} =128</tex> , although the encoding delay is as much as 128 ms, the bandwidth expansion is only about 100 Hz (using the new segmental measure), and the digit intelligibility <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">I</tex> is 20 percent. Note that in the specific problem of recognizing ten digits, purely random (input-independent) listener responses correspond to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">I = 10</tex> percent.

Electrical stimulation of the cochlear nucleus in a human subject

A patient requiring bilateral acoustic neuroma removal was implanted with electrodes in the ventral cochlear nucleus. Electrical stimulation was performed over six weeks. Thresholds for sinusoids from 50 to 4000 Hz were ⩽10 μA; frequencies above 10 kHz required 100 to 150 μA. Discomfort levels required about 35 μA at low and mid frequencies and 300 μA at high frequencies. Weber fractions for intensity were approximately 3% above 500 Hz and 17% at 90 Hz. Loudness growth functions for 250, 1000, and 2000 Hz stimuli were linear with a doubling of perceived loudness for each 3 to 4 μA increase in intensity. At 90 Hz, an accelerated loudness function was obtained. Experiments with direct analog speech stimuli suggested nasality and place of constriction as features used to discriminate consonants and vowels. The patient's sound perceptions with a House implant stimulator will be discussed.

Digital echo cancellation for baseband data transmission

In analogy with analog speech transmission, simultaneous two-way ("full-duplex") transmission of data signals over two-wire circuits can, in principle, be achieved by the application of hybrid couplers. In practice the resulting imperfect isolation between transmitter and receiver at the same end of a connection is disastrous for the detection of the data from the other end and has to be nullified in some way. A solution to this problem can be found in echo cancellation by means of an adaptive digital filter. In this paper an integrated circuit realized in 4Φ-dynamic NMOS technology is presented which has been designed especially for echo cancellation in baseband modems. In the design a number of aspects have extensively been accounted for. These are the choice of adaptation algorithm, the interaction between the digital filter and its analog environment, interpolation, finite precision implementation, and the natural occurrence of an effect akin to dithering. These subjects receive full attention in this paper. The realized IC can be used for the economical upgrading of existing baseband modems to full-duplex service and can pave the way to the use of similar circuits in the local area of the future digital telephone.

Analysis of Buffer Behaviour for Multiplexing of Data with Burst-Poisson Arrivals in Analog Speech Signals

A queueing model with finite waiting-room, burst (batch) Poisson arrivals, synchronous transmission and server interruptions through a first-order Markov process is studied. Using average burst-length and traffic intensity as parameters, the relationships among overflow probabilities, buffer size and expected burst-queueing delay due to buffering are obtained. A recursion method for computation of steady-state probabilities of the buffer states is developed. This allows an estimation of the buffer lengths required for overflow probabilities less than or equal to 10-6 with the use of lesser computer memory. A system of multiplexing of data with burst Poisson arrivals in analog speech signals is considered as an application of this model. The results of this study are portrayed on the graphs and may be used as guidelines in buffer-design problems in many digital voice-data systems. Although this problem arose in the study of data multiplexing in speech, the queueing model developed is quite general and may be useful for other industrial applications.

A microcomputer-based speech synthesis-by-rule system

This paper describes a system for generating speech by rule from a phonetic representation, using a resonance analogue speech synthesizer driven by parameters which are computed in real time by a microcomputer. Input to it, in the form of a phonetic transcription with additional markers to control rhythm and intonation, can come from a terminal attached to the microcomputer, or from a host computer linked to it by a serial line. A novel feature of the implementation is the use of a high-level structured programming language, “C”, which is compiled on the host and transmitted to the microcomputer as object code. This allows changes to be made quickly to the segmental and suprasegmental synthesis routines, which are still under development, and combines flexibility with ease of use in man-machine applications requiring speech output. To the host, the microcomputer/synthesizer system is simply a character-oriented low-data-rate output device.

Simulation model of buffer behavior for data multiplexing with Erlang arrivals in analog speech

Due to great simplicity in mathematical analysis, data-arrival process is generally assumed to be a Poisson process but it does not fit in many practical situations. However, the arrival process can be better approximated by Erlang distribution having appropriate degree. This paper analyses buffer behavior in queueing system with finite waiting-room, Erlang arrivals, synchronous transmission and single server having interruptions through a first order Markov process. Erlang distribution for arrival process puts multiple dependence on buffer states. Also this interruption process gives two simultaneous linear matrix equations. Thus the solution of state equations is difficult. Therefore, computer simulation has been used to study the relationships among overflow probability, buffer size and expected queueing delay due to buffering as a function of traffic intensity and the degree of Erlang distribution (Erlang parameter). A voice-data system (shown in Fig. 1) has been considered as an application for this model. The results of this study are portrayed on graphs and may be used as guidelines in buffer design problems in similar situations. Although this problem arose in the study of multiplexing of data in speech, the queueing model developed is quite general and may be useful for other industrial applications.

An on-off single server queueing model with finite waiting room and its application to computer communications

A queueing model with finite waiting room, Poisson arrivals, constant service time, synchronous transmission and a single server interrupted through a first-order Markov process is studied. The buffer's access to a channel, transmitting data at a field rate is controlled by a switch which in turn is controlled through a first-order Markov process. The relationships among overflow probabilities, buffer size and expected queueing delay due to buffering are obtained. A recursive method for computation of steady-state probabilities of the buffer states is developed. The system of multiplexing data in analog speech signals on telephone lines is considered as an application of the model studied. The results of this study are portrayed on graphs and may be used as the guide lines for the buffer design in digital systems. The system of multiplexing data in analog speech signals using this model, is simulated on the EC-1030 computer to check the validity of the analytical results. Simulation results are also portrayed on the graphs for comparison. Although this study arose in the buffer design required for multiplexing the data in analog speech, the queueing model developed is quite general and may be useful for other industrial applications.

Microcomputer-based system for speech synthesis from a large English text data base

The basic goal for the described project is selective and interactive production of synthetic speech from a large data base such as a training manual. In a training environment at a sophisticated computer control console, an operator should learn more effectively and efficiently if both aural and visual senses are used and if his eyes and hands are free to concentrate on the panel before him. The system described includes an operator-directed control strategy to selectively retrieve that part of text which is to be spoken, a considerable amount of text preprocessing in non-real time to obtain synthesizer control parameters, and a microcomputerized cascade terminal analog speech synthesizer. The preprocessing of the English text utilizes an extensive set of translation rules, dictionary lookup of complete words, and morph lexicon lookup of parts of words to produce 123 distinct allophone designators with corresponding stress, pitch, and duration designators. Storage requirements for the preprocessed speech control parameters to be supplied to the real-time speech synthesized are approximately 120 bits/sec of synthesized speech. Example recordings of the synthetic speech produced by this system will be presented. [Work supported by the Naval Ocean Systems Center.]

The Telephone Enquiry Service: a man—machine system using synthetic speech

The Telephone Enquiry Service is a computer system which allows interactive information retrieval from an ordinary touch-tone telephone. For input, the caller employs the touch-tone keypad, and the computer replies with a synthetic voice response. The service has been in fairly continuous operation for around one year, using a small time-shared computer in conjunction with an internal 200-line telephone exchange, and has been widely used by people with no special interest in synthetic speech. An unusual feature of the system is that the speech is generated by rule from a phonetic representation. A satellite computer, acting as a peripheral to the main machine, performs this task in real time, and controls the parameters of an analogue speech synthesizer. This constitutes an extremely economical and flexible method of speech storage, whose only real disadvantage is the low quality of articulation of the output. A major conclusion of the work is that even low-quality speech is acceptable to casual users, if the service is sufficiently interesting and useful to them.

Cascade/parallel terminal analog speech synthesizer and a strategy for consonant—vowel synthesis

A FORTRAN subroutine has been developed for the synthesis of speech and speech-like waveforms appropriate for psychophysical experimentation. The program accepts, as input, a set of control parameter values every 5 msec and produces, as output, samples of a synthetic speech waveform. Formant resonators simulating the transfer function of the vocal tract may be configured in cascade and/or in parallel (depending on the requirements of the user) by changing the value of the logical switch. There is a minimal change in the output speech when the switch value is changed. The parallel version of the synthesizer is both easier to control and a better model of the vocal tract transfer function than a number of earlier parallel synthesizers. A FORTRAN listing of the synthesizer is available upon request. Soon to be completed is a manuscript that describes simplified control strategies for the synthesis of various consonant—vowel syllables of English. [Work supported by NIH Grant NS 04332.]