Search

The conversion of speech into on-off pulse patteras representing talk spurts and pauses is a major tool used by psychologists in analyzing interactive communications behavior. We describe a conversion method and a means for interfacing its output with a computer, which allows the collection of large amounts of speech data from one or more sources. The major improvement over prior art is the ability to reject cross talk between neighbor speakers while permitting the detection of simultaneous speech. In the evaluation of the performance of the apparatus, no errors were detected for single-party speaking. Infrequent small errors occurred in the detection and duration of talk spurts and pauses during two- and three-party speaking, but these would probably be acceptable in most applications.

A sliding-scale coder for television signals was built which extends the range of the quantizing scale by processing the input signal twice when the input signal exceeds a prescribed threshold. On the second pass the quantizing range is effectively moved outward to reduce the errors in coding large signals. Double processing nearly triples the number of quantizing levels of a basic three-bit coder. Measurements of the number of extra bits required, that is, those in excess of three-bits per sample show that they may be accomodated on a three-bit per sample transmission channel by reducing the sampling rate five percent. The experimental coder generates 19 quantizing levels. Its performance approaches that of a seven-bit pulse code modulation coder. Busyness or streaking, common to most three-bit differential type coders, is eliminated. Acceptable pictures are reproduced with ±5 dB changes in the input signal's range. Over this range the signal-to-noise ratio of the reproduced pictures varies from 47 dB to 54 dB and the rise-time of a regenerated step-signal varies from 1 microsecond to 1.45 microsecond when the input signals rise-time is limited to 1 microsecond.

A subjective comparison of line-interlaced television pictures and noninterlaced television pictures has shown that the line-interlacing of low-resolution television pictures provides a bandwidth saving of considerably less than 2:1 when the line structure of the picture is visible. A line-interlaced television picture was subjectively compared with several noninterlaced television pictures in an effort to determine their subjective equivalency in terms of bandwidth. Several other variables — noise, spot-wobble, line-width to line-pitch ratio, different models, illumination and luminance — were also employed in the experiments. The televised pictures consisted of a head-and-shoulder view of a model pantomiming a two-way conversation. The results indicate that the line-interlacing of low-resolution television pictures provides about a 37 percent saving in terms of bandwidth at a relatively low value of high-light luminance of 40 fL (140 cd/m2) and as little as a 6 percent savings at a high-light luminance of 100 fL (340 cd/m2). When the line-width to line-pitch ratio is set at its preferred value for all pictures, a significant difference is obtained when the high-light luminance is decreased from 60 fL (200 cd/m2) to 40 fL (140 cd/m2). The effects of Gaussian noise, spot-wobble, illumination, and different types of models did not alter the subjective equivalence of line-interlaced and noninterlaced television pictures significantly. The addition of noise to a spot-wobbled picture was found to be more detrimental to the quality of the noninterlaced pictures than to the line-interlaced picture.