A physiological front-end preprocessor for speech recognition was evaluated using a large isolated-word database in quiet and noise. The front-end was based on the ensemble interval histogram (EIH) model developed by O. Ghitza. This model provides phase or synchrony information similar to that available on the auditory nerve. A modified EIH front-end was implemented and was tested using the Lincoln robust hidden Markov model isolated-word recognizer with a multistyle database at various signal-to-noise ratios (SNRs). The modified EIH front-end performed as well as a conventional mel-filter-bank front-end for normal speech. It provided a slight improvement in error rate at very low SNRs but required substantially more computation than the mel-filter-bank front-end

Algorithms that are implementable by artificial neural networks show promise of augmenting the field of automatic speech recognition. A specific approach to the problem of isolated-word recognition was initiated by Tank and Hopfield. These ideas were applied to a concatenated systems consisting of a vector quantizer, a time concentrator with vector sequences as input and allophones as output and a final stage with allophone sequence as input and isolated words as output. Improvements in the system are discussed; included are the addition of data-dependent 'phenomenological' rules that yield improved results for a single-speaker 35-word vocabulary isolated-word recognition task.>

HF speech communications is possible beyond the line of sight because signals in the 5-30 MHz range are refracted by the ionosphere. The fading these signals suffer and the channel impulse burst noise can be mitigated by a variety of channel equalizer and error correcting coding schemes. If the fade or burst noise is severe enough or long enough, however, speech communication becomes difficult or impossible. This paper describes how vocoders can be designed to overcome these difficulties in a way that is more or less transparent to the users, i.e., the users can talk through fades and not have to repeat themselves. Two methods are discussed; first, variable rate vocoding techniques and second, "speeded-up" speech. The speeded-up speech works as follows. During a deep fade, the transmitter retains the necessary speech parameter information. When the channel returns, this information is transmitted and transformed into speech at the receiver wherein the speaker is perceived to be speaking more rapidly, although with the correct pitch and spectrum. This allows the listener to "catch-up" with the speaker. A demonstration will be given to show how the vocoder performs under several simulated fading situations.

Preliminary results are reported on a pattern-matching vocoder experiment. In contrast to other reported approaches to this problem, the method employed involves a great deal of perceptual feedback from the experimenter, who incorporates patterns into the pattern set from a new sentence by first listening to the sentence as processed through the complete system. By identifying those portions of the processed sentence that have been degraded, he can augment the pattern set with only those frames that were degraded. The preliminary results indicate that a system with 500 patterns plus volume normalization, resulting in a 550 bits per second rate for spectral transmission, yields reasonably good synthetic speech.

Formant tracking and synthesis from formant information has been applied to a channel vocoder. The speech spectrum is first estimated with a 29 channel vocoder analyzer. Four crude formant values are obtained by processing the analyzer outputs. In the serial formant approach bandwidths and center frequencies for each formant are then computed by parabolic interpolation of neighboring channels. After normalization, bandwidths are recomputed so as to give a better match between measured and derived formant amplitudes. Finally, a spectrum derived from four formant frequencies, four formant bandwidths plus a single gain factor replaces the measured spectrum and the speech is synthesized by the spectrally flattened channel vocoder synthesizer. In the parallel formant approach, four formant frequencies, four formant amplitudes plus the amplitudes of the first and last channel are transmitted. The remaining channel signals are computed at the receiver by parabolic interpolation. Preliminary listening results indicate that these representations can be effective in reducing the bandwidth of a channel vocoder.

Integrated packet-switched networks have potential for providing improved performance by dynamically sharing transmission bandwidths between various users and user types, but new flow control methods are needed to deal with packetized voice traffic. This paper describes a packet voice flow control concept based on embedded speech coding. Results are presented from a computer simulation study of the technique in the context of a multilink wideband packet speech network. Several control methodologies are described, leading to an end-to end feedback approach that achieves stable operation and efficient utilization of network resources by adaptively matching transmitted voice bit rates to prevailing network conditions. Issues in the design of embedded speech coding algorithms are reviewed and a candidate structure based on channel vocoding principles is presented, along with the subjective results of some preliminary listening tests

This paper contains a description of the architecture of the fast digital processor (FDP), a general purpose digital attachment to a UNIVAC 1219 computer facility. The main purpose of the FDP is to enhance the capability of the UNIVAC facility for performance of digital signal processing operations such as digital filtering and discrete Fourier transforms. The structural design evolved during a series of discussions among the four authors. The FDP is presently being constructed under the supervision of one of the authors, P. McHugh, and should be completed in late 1970.

A problem which arose during research involved in designing a machine to translate hand-keyed Morse code into printed text may be stated as follows: LetX = \{x_i: i = l, 2, \cdots , n\}be a sequence of independent random variables all of which have the same distribution. Assume that the probability thatx_i = x_j,i \neq j, is zero. Letkbe a positive integer\leq n, and consider all subsequencesx_i, x_{i+1}, \cdots , x_{i+k-1}ofXconsisting ofkconsecutive variables. Let us distinguish, with a check (\surd), the largest member of each such subsequence. We have studied, and partially tabulated,p_n^k (r), the probability that exactlyrmembers of the sequenceXare not checked. This paper contains most of the pertinent results.

A transistorized special purpose digital computer called MAUDE (Morse AUtomatic DEcoder) has been designed, built and analyzed. This computer decodes a hand-sent Morse message, which is printed on a teletypewriter. MAUDE's decisions take place at a number of different levels and its "knowledge" is not only that of relative durations of dots and dashes, but also of the Morse code and even of certain elementary properties of language. MAUDE has successfully decoded between 90 per cent and 95 per cent of 184 operators. A successful decoding is one which results in a text which can be easily read by a man who knows the language. It is felt that MAUDE can be a practical piece of equipment for a site with heavy traffic. Its performance will be inferior to that of a man until more sophisticated language rules, using at least a word vocabulary, are included.