Telephone Quality Research Articles

Subjective quality quantification of digital voice communication systems

The expanding digitisation of the telephone networks has brought about an increasing interest in methods addressing the performance characterisation of new digital voice transmission systems. In the paper a review of subjective methodologies for the assessment of telephone quality digital voice communications systems is provided. Technical aspects concerning the network applications and other characteristics relevant to the type of system under evaluation are considered first, since these factors influence the selection of a suitable assessment methodology. Next, listener opinion and conversational opinion tests are described, with particular emphasis on their respective merits, characteristics and differences. Finally, articulation and diagnostic tests, while not generally employed for the assessment of telephone quality, are briefly examined, since the results obtained from such tests are often correlated with the perceptual performance of telephone quality communications links.

An improved word detection algorithm for telephone quality speech

Accurate location of the endpoints of spoken words and phrases is important for reliable and robust speech recognition. The endpoint detection problem is fairly straightforward for high‐level speech signals in low‐level stationary noise environments (e.g., signal‐to‐noise ratios greater than 30 dB). However, this problem becomes considerably more difficult when either the speech signals are too low in level (relative to the background noise), or when the background noise becomes highly nonstationary. Such conditions are often encountered in the switched telephone network when the limitation on using local dialed‐up lines are removed. In such cases the background noise is often highly variable in both level and spectral content due to transmission line distortions, transients, and tones from the line and/or from signal generators, etc. Conventional speech endpoint detectors have been shown to perform very poorly (on the order of 50% word detection) over these conditions. In this talk we present an improved word detection algorithm which incorporates both vocabulary and task information, leading to word detection accuracies close to 100% for isolated digit detection over a wide range of telephone transmission conditions.

Transmission loss planning for digital networks

AbstractThis paper describes the transmission loss plan for a digital network on the basis of transmission quality evaluation data for telephone, modern and facsimile services. First, reference equivalent distribution of a subscriber loop and terminal return loss distribution at 2‐wire/4‐wire conversion circuit of digital local exchange subscriber circuit, which are network parameters for transmission loss planning, are estimated based on investigations of the actual conditions. Next, with respect to the telephone, the relation between transmission loss and opinion is clarified both for a short length of connection where degradation due to echo is negligible and for medium and long length of connections where the echo performance is taken into account. With respect to modem and facsimile, the relation between openloop loss and quality degradation is evaluated experimentally. As a result of total evaluation of network parameters and evaluation data with respect to telephone, modem and facsimile, it is revealed that the transmission loss in EO‐EO all‐digital network should be set at 4 dB for intra‐office connecnection. Here, degradation due to echo is negligible and 8 dB for medium and long length of connections where echo performance is taken into account to maintain good telephone and modem‐facsimile transmission quality. Finally, the transmission loss plan from existing analog network to EO‐EO all digital network is shown.

A study of time-encoded speech

The technique of Time-Encoded Speech (TES) has shown considerable promise in transmitting good quality speech at bit rates below 16 kb/s. The central concept of this approach depends upon the transmission of coded shape descriptors for successive extended segments of the speech waveform. For transmission over a constant rate data channel, buffer management via a first-in first-out store is required. A modified time encoding method for speech signals which is basically a kind of extremal coding is described. This technique enables the low frequencies of speech waveform to be reconstructed more faithfully, and this is desirable for human perception. With linear encoding and transmission rate at about 12–16 kb/s, the TES output speech quality approaches that of public telephone quality, and little improvement can be gained by further increase in the data rate.

Voiced/unvoiced band-switching system for transmission of 6 kHz speech over 3.4 kHz telephone channels

A voiced/unvoiced band switching system is described for the transmission of wide bandwidth speech (here 300to 6000 Hz) over commercial telephone channels. Timewaveforms and spectrograms are presented to demonstrate that the 6 kHz speech can be compressed into a 3-4 kHz telephone channel and yield improved speech quality at the receiver compared to current telephone quality speech. Informal listening experiences confirm these results. The system is conceptually simple and inexpensive to implement.

Speaker-independent recognition of isolated words using clustering techniques

A speaker-independent isolated word recognition system is described which is based on the use of multiple templates for each word in the vocabulary. The word templates are obtained from a statistical clustering analysis of a large database consisting of 100 replications of each word (i.e., once by each of 100 talkers). The recognition system, which accepts telephone quality speech input, is based on an LPC analysis of the unknown word, dynamic time warping of each reference template to the unknown word (using the Itakura LPC distance measure), and the application of a K-nearest neighbor (KNN) decision rule. Results for several test sets of data are presented. They show error rates that are comparable to, or better than, those obtained with speaker-trained isolated word recognition systems.

An integrated interpolative PCM decoder

Discusses the technique of time averaging interpolation which provides a means for obtaining telephone quality digital-to-analog conversion using conventional digital integrated-circuit processing. Describes a monolithic circuit that decodes the 8-kHz companded PCM commonly used in voiceband communication systems. The circuit, realized in a standard buried collector bipolar technology, contains 300 integrated injection logic (I/SUP 2/L) gates together with a current-driven resistive ladder network. The performance of the decoder exceeds typical toll network objectives.

An Integrated Interpolative PCM Decoder

The technique of time-averaging interpolation provides a means for obtaining telephone quality digital-to-analog conversion using conventional digital integrated-circuit processing. This approach has been adopted in a monolithic circuit that decodes the 8-kHz companded PCM commonly used in voiceband communication systems. The circuit, realized in a standard buried collector bipolar technology, contains 300 integrated injection logic <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">(I^{2}L)</tex> gates together with a current-driven resistive ladder network. The performance of the decoder exceeds typical toll network objectives.

Real-time harmonic pitch detector

A real-time harmonic pitch detection algorithm has been developed on the Lincoln Digital Voice Terminal (LDVT). The algorithm was designed to be fast and to perform well when the input speech is degraded (i.e., telephone quality) or corrupted with acoustically coupled noise. The algorithm determines the fundamental frequency from the spacing between harmonics in a selected portion of the spectrum. The algorithm was incorporated into a real-time linear prediction vocoder and compared favorably in informal listening tests with the Gold-Rabiner time-domain detector under a variety of adverse conditions.

Performance of Harpy speech recognition system for telephone quality speech input

Spectral distortions in speech affecting the performance of speech processing systems can be classified into multiplicative and additive types. We report in this paper the results of our study on the performance of Harpy continuous speech recognition system for multiplicative type of distortion with special reference to telephone quality speech. It has been observed that the recognition performance of Harpy system for digit task dropped from 99% to 93% for telephone speech [G. Goodman, B. Lowerre, D. R. Reddy, and D. Scelza, J. Acoust. Soc. Am. Vol. 60, S11(A)]. The effect of distortion on segmentation and labeling will be discussed for simulated and actual telephone data. Performance evaluation is obtained for A1 information retrieval task (1011 words) in terms of word and sentence accuracies for different types of distortion. The results will be discussed with reference to signal processing functions in the system and some normalization techniques are suggested to account for the distortion. Comparison ...

Real-number convolutional codes for speech-like quasi-stationary sources (Corresp.)

A quasi-stationary source is one which is stationary over short periods, but changes occasionally to a new mode of stationarity; a typical example is the speech source. Convolutional codes are designed for the speech source for use with search algorithms such as the Viterbi, stack, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">M</tex> -algorithms; the design is first by heuristic means and then by an empirical optimization. These codes have the same generating structure as the usual binary convolutional codes, but employ ordinary arithmetic. It is found by experiment that, at rate 2 bits/sample (which allows telephone quality speech), such codes need not have constraint length longer than five or six, which implies a generating circuit of about 1000 states. Encoding noise becomes white and uncorrelated with simple code searching; this shows that a short fixed code can successfully decorrelate waveforms from different source modes.

Adaptive Delta Modulation with a One-Bit Memory

We propose a delta modulator which, at every sampling instant r, adapts its step-size (for a staircase approximation to the input signal) on the basis of a comparison between the two latest channel symbols, C r and C r-1 . Specifically, the ratio of the modified step-size m r to the previous step size m r-1 is either +P or — Q depending on whether C r and C r-1 are equal or not. (We recall that, in delta modulation, C r represents the polarity of the difference, at the sampling instant r, between the input signal X r and the latest staircase approximation to it, Y r-1 .) A simulation of the delta modulator with a band-limited speech input has revealed that PQ = 1 and P ⋍ 1.5 represent optimal adaptation characteristics, on the basis of signal-to-error ratios, over an important range of sampling frequencies; and that at 60 kHz, delta modulation with these adaptation parameters compares favorably with 7-bit logarithmic PCM, which reproduces speech with good telephone quality. We present several graphical results from this simulation, and include an evaluation of the effect of independent channel errors on the adaptive delta modulator. We proceed to suggest a heuristic theory of the delta modulator which explains the optimality of the condition PQ = 1, and develops an upper bound of 2 for the optimum value of P. We conclude with a summary of results from a video simulation which revealed that aforementioned optima for P and Q apply to a video signal

Project for a Digital Telephone Network

A digital communication network is proposed that would integrate present telephone services and switched mediumspeed data or facsimile channels into a single general-purpose system. Efficient use of the cable plant is made through decentralization of switching and centralized control. Subscribers are connected to voice or data concentrators, and switching between different digital trunks is done in standardized asynchronous switching units. A novel synchronization method prevents any information losses. In view of the optimization of such a system, a reappraisal of basic system characteristics has been undertaken. For example, a ternarybased code system is used together with dc-balanced binary transmission. This facilitates the construction of simple noncritical terminals and repeaters. To allow all sorts of digital speech processing, such as conferencing or digital filtering, a companding law is chosen such that a linear addition of compressed samples is easily realizable. A higher than usual sampling rate of 12 kHz was chosen for several reasons: 1) simple low-pass filters in the PAM voice-concentrator are possible; 2) it allows a potential voice bandwidth of 6 kHz with a corresponding increase in telephone quality; and 3) data or picture transmission speed is considerably increased.

Detection of Noiselike Sounds in Telephone Speech

A method of separating noiselike sounds from vowel/vowellike sounds where the speech is of telephone quality is presented. This procedure involves the use of the χ2 nonparametric test, which analyzes the fluctuations of the time waveform and separates the speech into two classes depending on the randomness of the waveform. Initial computer simulations show the feasibility of the method, but since the recognition networks must operate in real time and the χ2 test is nonlinear, a linear model to the χ2 test is constructed that can be easily implemented, operate in real time, and obtain an efficiency close to that of the χ2 test. The false-alarm and false-recognition rates obtained using the linear model on a test sentence of 45 phonemes and 6-sec duration are in the order of 12% and 85%, respectively.

A Modern Version of the Old Theory of Telephone Quality

In 1921 a method was formulated for calculating the articulation—the percent of speech sounds interpreted correctly during a telephone conversation—from the physical characteristics of the telephone system and the noise environment of the listener. This paper describes this empirical theory as recently revised to correspond to modern techniques. A quantity A is calculated which is called the articulation index and which is related to the various kinds of measured values of articulation. In particular, if s is the percent of the fundamental sounds correctly perceived by the listener A = −p log (l-s), where p is a factor depending upon the skill of the crew making the articulation tests. For a well-practiced crew the value of p is between 0.5 and 0.6. The value of A is calculated by the equation A = F⋅V⋅E, where F is the maximum value of A, and V a factor which is zero at threshold intensity of the speech received by the listener and grows as the received speech intensity increases reaching unity for intensities between 60 and 70 db above threshold. The factor E is unity below this intensity but decreases from unity to 0.85 as the intensity levels increase to 60 db above these levels. The factor F is calculated by the equation F = ∫0∞D⋅Wdf, where D is a function of the frequency f, and W is a function of the relative response of the system. The functions D and W have been determined directly from articulation data. The factors V(x) and E(x) are determined directly from articulation data. The value of x for systems in a quiet place is equal to a weighted average of the response of the system. When a noise is present which produces a masking M at each frequency at the listener's ear, the same functions are used for determining F and V, but the response R at each frequency is replaced by R − M. However, for this case the x in E(x) is increased by an amount Δ, a quantity calculated directly from the masking M of the noise and the intensity of the received speech.

Effects of Phase Distortion on Telephone Quality*

This paper discusses the effects of the type of phase distortion found in low-pass filters and the loaded line on telephone quality. The effects are ascribed to three factors; the first involves the slopes of the phase characteristic at various frequencies in the range of interest, the second involves the intercept values on the phase shift axis of the tangents to the phase curve, the third involves the interference caused by portions of one sound overlapping portions of a succeeding sound. The first factor appears to be the one of most importance.