Alan Carpenter

Judgement of the Vowel Colour of Natural and Artificial Sounds

Five vowel sounds, each produced in five different ways, were presented to subjects with widely differing experience of tasks of this type for identification. The vowels were the last five of the standard set of eight cardinal vowels, and were presented to subjects with no experience of phonetics as belonging to the words bat, bart, bought, boat and boot. Each vowel was (a) spoken by a phonetician, (b) synthesized by adding individual harmonics in direct imitation of the spoken sounds, (c) represented by two pure tone components only, (d) synthesized by a talking machine (P.A.T.) and (e) imitated by selecting harmonics with a low-pass filter. All these sounds were categorised more or less “correctly” by skilled subjects; and, although there were wide and systematic differences in accuracy with untrained subjects, these performed consistently better than chance. The experiment showed (a) that the selection of harmonics according to formant theory is not the only, and perhaps not always the best means of synthesizing isolated vowels, (b) that even sounds which are poor representations of vowels can still be categorised with some consistency even by untrained subjects and (c) that a simple “percent correct” score is less sensitive than other measures.

Comparisons of Training Techniques for Complex Sound Identification

Fourteen different synthetic engine sounds were mixed with five different synthetic cavitation sounds at realistically difficult signal‐to noise ratios. Eighty one Royal Navy ratings, divided into eight training groups, were asked to identify the 14 engine sounds. Training procedures were varied among groups to answer some questions pertinent for programming a teaching machine. It was found that (1) verbal descriptions of the actual physical characteristics of the sounds give better results than those supplied by experienced listeners, (2) the order in which training items are presented is important. The most useful order involved changing one relevant quality per item, (3) “good” high engine‐to‐cavitation ratio) recordings may be an advantage if alternated with realistically “bad” ones, (4) feedback (knowledge‐of‐results) procedures should be concentrated more at the end of a training program and recordings should be extended temporally to overlap with the feedback, (5) a large population of engine noise...

Experiments Relating to the Perception of Formants

Experiments are described which suggest that, for human perception, the information concerning the location of a formantlike complex sound is contained in the two most prominent harmonics. This result is limited to the condition where adjacent harmonics are more widely separated than the width of a critical band.

Perceptual Confusions between Four‐Dimensional Sounds

Are meaningless sounds learned and identified in a speech mode? Sixteen sounds were generated differing on four dimension, each having two values. These were: the source waveform contained either all harmonics or odd harmonics; the fundamental frequency was either 90 or 142 Hz; there were either one or two formant frequency regions; and the formant(s) was (were) either low, 600 (and 1550) Hz or high, 940 (and 2440) Hz. Twenty‐four listeners identified these sounds. Fewer confusions were made between sounds (1) as the number of dimensions on which they differed increased, and (2) as the dimension(s) changed, from formant number to ferment frequency region, to fundamental frequency, to source waveform. Similarly, fewer confusions were made between sounds whose formant frequencies and fundamental frequency were in a fixed ratio than between those with the same formant frequency. In essence fixed ferment sounds were more often called the same sound than were fixed ratio sounds. The data fit a hypothesis that ...

The Perception of Vowel Colour in Formantless Complex Sounds

Complex sounds were manufactured by passing the output of a buzz source through band-pass filter networks. The resulting stimuli were played both to phoneticians and to phonetically naive subjects who were asked to judge the vowel colour of the sounds. The responses showed that complex sounds are categorised consistently with regard to their vowel colour in spite of the absence of formant peaks. The following additional conclusions can be drawn: (1) The judgement of vowel colour is relatively unaffected by such features of the stimulus as variability in length between 300 and 700 msec. and the temporal qualities of the on-off transitions. (2) Vowel sounds of the back group can be simulated with a low-pass filter; and increasing the cut-off frequency shifts judgement in the direction from “high” to “low”, i.e. from /u/ to /α/, and then /a/.

Discrimination of Vowel Sounds Synthesized By Harmonic Addition

An experiment is described in which synthetic sounds were identified by untrained subjects as vowels. The sounds were constructed by adding together harmonics selected from a buzz source in the same relative amplitudes as were found in corresponding natural vowels. This work is the first step in an attempt to identify those qualities of a complex sound which determine its perception as a particular vowel, irrespective of theories based upon the mechanism by which speech sounds are produced.

Identifying Meaningless Tonal Complexes

Series of buzz tones with up to 24 harmonics were presented to five groups of listeners for identification. Nine tones in which different harmonics were emphasized were presented to Group I who could easily tell them apart and could identify them with 33% accuracy. The remaining groups heard three of the tones in various conditions of noise, filtering, fundamental frequency, and emphasized to nonemphasized harmonic intensity differentials. The most difficult task was to identify the tones with differing harmonic structure when the fundamental frequency was not the same for every complex tone. It was also difficult to pick out complexes with the same harmonic structure and basic frequency when noise‐masked, filtered, and in‐quiet items were intermixed at random in the same test. The distinguishing differences in the meaningless buzz complexes were in harmonics six and above.

Identifying diesel engine sounds

Abstract Naval ratings were given the task of identifying the sounds of three lorry diesel engines at running speeds of 1000, 2000 and 3000 rev/min. The physical signal was varied by adding thermal noise and by multiplying all frequencies by factors of 1, 2, 4 and 8. The experimental variables were knowledge of results, amount of cueing, and the meaningfulness of the identifying names. Meaningfulness consisted either of code names or of the real names plus a description of the physical characteristics of the engine. Where real names were in use, identification was more accurate even with frequency multiplication. Engine types were identified better than engine rev/min. The accuracy of engine identification decreased as the noise masking increased and as the frequency multiplications changed from × 1 to ×2 to ×8 to ×4. However, when noise masking was absent, engine identification at the ×8 speedup yielded scores equal to ×1 and ×2. The rev/min identifications were all equivalent except at ×4 in noise, which always gave the lowest score.