Speech Interference Research Articles

Segmental Intelligibility and Speech Interference Thresholds of High-Quality Synthetic Speech in Presence of Noise

Technological advancement in the area of synthetic speech has made it increasingly difficult to distinguish quality of speech based solely on intelligibility scores obtained in benign laboratory conditions. Intelligibility scores obtained for natural speech and a high-quality text-to-speech system (DECtalk) are not substantially different. This study examined the perceived intelligibility and speech interference thresholds of DECtalk male and female voices and compared them with data obtained for natural speech. Results revealed that decreasing signal-to-noise levels had more deleterious effects on the perception of DECtalk male and female voices than on the perception of natural speech. Analysis of pattern of phoneme errors revealed that similar general patterns of errors tended to occur in DECtalk and in natural speech. The speech interference test did not demonstrate any significant difference between the DECtalk male and female voices. These results were supported by the absence of a significant difference between DECtalk male and female voices during intelligibility testing at different signal-to-noise ratios.

A Survey on Noise Exposure Level in an Aluminum Can Production Plant.

The present investigation was designed to quantify noise exposure of laborers at various sections of an aluminum can production plant. The acoustic environment in each area was described by sound pressure level (SPL). Decibel (dB) readings were taken for a range of individual frequencies and weighting scales (A and C). Speech interference level (SIL) values for laborers in different production sections were also determined. Moreover, by means of a personal noise dosimetry technique, the equivalent continuous noise exposure level (dB(A)) during an 8-h shift (Leq, 8) was computed. The results obtained were as follows. Spectral analysis of the noise indicated that high SPL associated with high frequency noise exceeded the permissible limit against hearing damage, and it generally amounted to levels higher than 90 dB(A), whereas those inside the control rooms were in the range of 54-60.5 dB(A). Assessment of the results obtained from personal noise dosimeters also revealed that the Leq,8 values were higher than the prescribed limit (85 dB(A)). By preliminary audiometric examinations on the right ears of workers, hearing threshold shifts (HTS) were noticed in the range of 7-11 dB at 1 kHz and 12-16 dB at 4 kHz. With regards to the SIL values (93.2-96.5 dB(A)) which were more than the maximum vocal effort, communication between workers and/or with supervisor were appraised to be very difficult (shouting) or impossible. It was concluded that (a) the dB readings observed in this survey indicate a potentially serious possibility of hearing disability, even though high level of HTS could not be noticed in the preliminary audiometric test, and (b) the risk of accidents should also be taken into account regarding the nonauditory effects of such a high level of noise. We suggest that (1) minimizing the transmission of noise through acoustical countermeasures need to be considered, and (2) introducing a schedule comprising on-time and off-time could be beneficial in a workers' health conservation program.

Noise control evaluation for control rooms and crew quarters on off-shore platforms

Noise sources on off-shore drilling and production platforms are similar to those found on-shore; i.e., engines, turbines, generators, gear boxes, pumps, compressors, etc. The major difference is they are located on a steel structure that transmits noise and vibration energy very efficiently. For those that live aboard, for a crew shift of as much as 3 weeks, their exposure time may not be limited to the 12-h daily work shift. Thus areas like control rooms, laboratories, bunk houses, dining, and recreation rooms should have lower noise levels, be located away from noise sources, and/or isolated from structureborne and airborne noise. This requires a different criteria and design approach. The criteria for this type of noise control problem is related to the speech interference and sleep interruption. Two case histories will be discussed that illustrate the scope of the problem, practical noise control solutions, and results. These include: (1) a semisubmersible combination drilling and production platform located in the Gulf of Mexico; and (2) smaller fixed leg flow station platforms off Indonesia.

Assessing the importance of subvocalization during normal silent reading

A concurrent speaking paradigm was used to assess the importance of subvocalization during the reading of lengthy natural prose passages. Experiment 1 showed that having subjects count aloud while reading interfered with their comprehension and recall of the text's details as well as its gist, but did not affect the durability of the memory trace. Experiment 2 replicated these findings and established the validity of using concurrent speaking as a technique to interfere with speech-specific processes during silent reading. By pitting concurrent speaking against a nonverbal concurrent task, Experiment 2 provided evidence that its detrimental effect on comprehension was due to a competition for speech-related resources rather than a general competition for cognitive resources. Interfering with speech recoding during silent reading led to an average decrement of 10–12% in comprehension performance. However, Experiment 2 also showed that there were substantial individual differences in the magnitude of the speech interference effect and that these differences were systematically related to subjective reports about the concurrent speaking manipulation.

Traffic noise exposure planning: A case application

The traffic noise situation around a projected extension of a motorway was evaluated by using the toxicological concepts of risk assessment. General annoyance, speech interference and sheep disturbances were identified as the major effects of the projected changes in the traffic pattern, which were to include decreases in the number of vehicles on three city streets and a larger proportion of heavy vehicles using the motorway. Provided that action was taken to reduce further the number of heavy vehicles on city streets, a decrease in the extent of annoyance was predicted. The increase in exposure around the projected motorway owing to increased maximum noise levels could be avoided by building noise barriers and introducing speed reductions.

Comparison of community reactions to traffic noise

In 1989, the community reaction to aircraft noise was surveyed around the New Tokyo International Airport (Narita Airport) by means of an interview method. The questionnaire used in the survey was a modified version of a question sheet used in a study around Yokota Airbase in 1971. The results of these two surveys were compared. The response to questions on interference with conversation, talking on the telephone and listening to TV and radio was quite similar in the two studies. However, some differences were observed in the rates of interference with night sleep and reading and/or concentrating. This may be caused by differences between the two studies in aircraft flight patterns and occupations of the respondents. Such tendencies were also observed when the Narita survey was compared with other studies conducted around the airports of Yokota, Haneda, Chitose, Osaka, Fukuoka and Miyazaki. Finally, the community reaction to aircraft noise was compared with the reaction to road traffic noise and train noise. The rates of annoyance and speech interference were highly dependent on noise levels. Other relevant factors, such as sex, age, year of residence, occupation, etc., had much weaker relations to the extent of reactions.

Noise Pollution in the Operating Theatre

Sound levels during a typical major operation were measured to identify the main sources of noise in the operating theatre. Although overall sound levels were within the recommended levels for a satisfactory working environment, loud intermittent noises of up to 108 dB were emitted from sources such as suckers, intercoms, and alarms on anaesthetic monitoring devices. The noisiest time was usually during the preparation period of the operation; during surgery, noise levels were much higher than levels of normal speech between staff. Preferred speech interference levels were often exceeded which made communication difficult and sometimes impossible. Communication and concentration were also disrupted by unnecessary background conversation.

Signal processing techniques for identifying sound quality parameters

Importance is now being placed on the type of sound that machinery makes rather than simply on the noise level it produces. Two reasons for this emphasis are: the adverse affect an unpleasant acoustic environment can have on workers' performance; and the link a consumer makes between the quality of a machine and the sound it makes. The current problem is to find objective measures that reflect this subjective quantity known as sound quality. It is important to define such a quantity based on objective measures so that design objectives or performance specifications can be formulated rigorously. Single number indicators that are in use now include loudness, speech interference level, sound‐pressure level (A‐E weightings), and impact noise ratings. In the automobile industry, aspects of sound quality such as muddines, roughnes, and linearity have been shown to be correlated with the relative power of harmonic components in a spectrum and also to the statistics of the time history envelope. However, these indicators can yield similar values for signals that sound very different and cause a different subjective response. In order to identify the relationship between different signal components and the sound quality rating, techniques that can alter particular components of signals are needed to generate playback signals for subjective testing. In this paper, signal processing techniques are described that have been used to vary noise signal parameters so that components that contribute to sound quality can be identified.

Noise pollution in the operating theatre

Sound levels during a typical major operation were measured to identify the main sources of noise in the operating theatre. Although overall sound levels were within the recommended levels for a satisfactory working environment, loud intermittent noises of up to 108 dB were emitted from sources such as suckers, "intercoms", and alarms on anaesthetic monitoring devices. The noisiest time was usually during the preparation period of the operation; during surgery, noise levels were much higher than levels of normal speech between staff. Preferred speech interference levels were often exceeded which made communication difficult and sometimes impossible. Communication and concentration were also disrupted by unnecessary background conversation.

Relationship between consonant recognition and subjective ratings of hearing aids

The relationships between consonant recognition performance and subjective ratings (using continuous discourse and nonsense syllables as stimuli) of speech understanding, listening comfort, quality, noise interference and overall liking were examined in quiet and in noise using hearing-impaired subjects and seven different hearing aids. Correlations among the subjective scales varied depending on test conditions and test stimuli. Scores on the speech understanding and overall liking scales correlated moderately with scores on a consonant-recognition test presented in noise and when syllables were used as the materials for judgments. The overall liking scale (for a hearing aid) correlated with all other subjective scales and consonant recognition performance. However, the relative contribution of each subjective scale to overall liking varied depending on test conditions and test materials.

Intelligibility of synthetic speech in the presence of interfering speech

Standard articulation tests are not always sensitive enough to discriminate between speech samples which are of high intelligibility. One can increase the sensitivity of such tests by presenting the test materials in noise. In this way, small differences in intelligibility can be magnified into large differences in articulation scores. We used both a more conventional articulation test and a monosyllabic adaptive speech interference test (MASIT) to evaluate the intelligibility of nine different speech-coding techniques. We found different patterns of responses for the articulation test and MASIT. These differences can be explained by the fact that different speech-coding schemes code different acoustic-phonetic properties of the speech signal. Some of these properties are more liable to masking by interfering noise than others. Our results show that, in the case of synthetic speech, differences in intelligibility are not always magnified by adding interfering noise: they may even disappear.

Evaluation of speech interference due to noise and reverberation

Experiments were conducted to investigate the effect of noise and reverberation on speech intelligibility and to find a simple method for evaluating the quality of speech transmission. Articulation scores and subjectively judged intelligibility of sentences were obtained in a laboratory situation together with the physical measure of the cross-correlation between the source and the listening point. The effect of broadband steady noise was examined by a monosyllable articulation test and subjective judgment of speech interference. A three-syllable articulation test was used to detect the effect of reverberation. These subjective indices showed correspondence with the source to listening point cross-correlation. The measurement of cross-correlation appeared to be effective as an objective method for estimating speech interference.

Activity disturbances caused by different environmental noises

An analysis was made of data obtained in investigations performed on annoyance reactions due to noise from aircraft, trains and road traffic. The aims of the analysis were to study the covariation between the extent of activity disturbances and general annoyance, to compare different noise sources with reference to activity disturbances caused, and to study the relation between the extent of annoyance reactions and noise exposure. The activity disturbances studied were speech interference effects, adverse effects on rest and sleep and awakening. Common to all three noise sources is that the extent of all activity disturbances increased with increased general annoyance. However, the rank order between the effects was different. In areas exposed to aircraft and train noise, speech interference was the effect reported by the largest proportion of respondents. Road traffic noise primarily disturbed rest and sleep. As regards train noise, it was demonstrated that although the extent of activity disturbances was comparatively high, the extent of general annoyance was lower on the whole than the extent of activity disturbances. These differences in responses may be due to differences in attitudes towards the noise sources. Even though actual noise-related effects were experienced, the degree of subjective annoyance may be determined by the type of noise source causing the effects. The dose-response relationship established for aircraft and train noise demonstrated that the extent of general annoyance varied although the noise conditions were similar. The differences were further accentuated with increasing dB(A) levels. The results from the analysis demonstrated that the use of response indices based upon different activity interferences must take into account the variations that are present between different noise sources. Weighting all activity interferences into an index and applying this universally for all noises will decrease the precision of the response description and hence influence the accuracy of the dose-response relationship.

Considerations of Intention and Movement: Action Plans and Motor Programs in Speech and Keyboard Entry

The cognitive organization and control of rapid movement sequences was investigated in speech and keyboard entry. Five reaction time experiments were carried out with various manipulations of stimulus meaning. A linear increase in reaction time with an increase in the number of items to be produced was evident for spoken lists of random digits, single words in random order, and repeated syllables but not for single words arranged into meaningful context. A speech interference condition produced a linear increase in reaction time in a key pressing task when it was necessary for the action to be planned ahead of time. These results indicate that meaningfulness of stimuli facilitates movement organization, an effect which can have significant impact on input technology.

Building noise criterion curves, BNC, for interior spaces

This paper presents an updated set of noise criterion (NC) curves that are renamed building noise criterion (BNC) curves. They are based, in part, on the latest definition of four‐band speech interference level and on “spectrum balance,” that is to say, the premise that the loudnesses of all bands containing the same number of critical bands should be equal, as calculated by Stevens Mk. VII perceived loudness method. The curves are extended downward in frequency to include the two octave bands with mean frequencies at 16 and 31.5 Hz. Finally, the high‐sound level, low‐frequency region of the curves [Blazier, Noise Control Eng. 16, 64–73 (1981)] are demarked to show where human annoyance will probably result from vibrations caused by such noise levels in contemporary building construction. The paper details the use of the BNC curves in writing a specification for building construction and in determination of the conformance of the measured result to the specification, or in rating an existing noise. Particularly important are the procedures given for determining “spectrum imbalance.” The handling of two typical types of imbalance are discussed: (1) an acceptable speech interference level accompanied by high low‐frequency band levels; (2) acceptable low‐frequency band levels accompanied by a very low speech interference level.

Effects of spectral alternation on the intelligibility of words and sentences.

In this study of a novel speech interference paradigm, it was found that periodic alternation between nonoverlapping filtering conditions (1100 Hz low-pass and 1700 Hz high-pass) significantly reduced intelligibility. Thus, although repetition accuracy for key words within sentences was approximately 98% with continuous low-pass or high-pass filtering alone, listeners averaged only 48% correct when frequency ranges alternated at a rate of 2.67 Hz. Interestingly, intelligibility improved by about 39% when low-pass noise was mixed with segments of high-pass speech and high-pass noise was mixed with segments of low-pass speech. A smaller increase due to added noise was also observed with spectrally alternated word lists. The interference produced by spectral alternation appears to be related to other verbal and musical phenomena, including stream segregation. The results from various control conditions indicate that the enhanced intelligibility produced by added noise involved phonemic restorations that occurred within each speech band, as well as a spectral completion effect that acted across spectral bands.

Temporal properties of memory for speech in preschool children.

The ability of four-year-old children to remember syllables was investigated, both to determine the temporal properties of memory for speech in subjects who do not use sophisticated retention strategies, and to determine what speech memory is available in the late preschool period. In Experiments 1 and 2, children were to remember pairs of nonsense syllables for 5, 10, 15, or 20 sec filled with a silent, manual task. In both experiments there was an apparent decay of one type of speech memory across 10-15 sec. In Experiment 2, strong interference was obtained when there was a final speech suffix item. An additional phase of this experiment verified that children cannot rehearse during the postlist period. However, in Experiments 3 and 4, suffix effects were not obtained with adult subjects in similar tasks. The results suggest that memory for speech persists for at least 10 sec even in preschool children, but that adults, unlike preschool children, engage in an additional type of speech coding that increases the resistance of memory to speech interference.

Activity interference and noise annoyance

Debate continues over differences in the dose-response functions used to predict the annoyance at different sources of transportation noise. This debate reflects the lack of an accepted model of noise annoyance in residential communities. In this paper a model is proposed which is focussed on activity interference as a central component mediating the relationship between noise exposure and annoyance. This model represents a departure from earlier models in two important respects. First, single event noise levels (e.g., maximum levels, sound exposure level) constitute the noise exposure variables in place of long-term energy equivalent measures (e.g., 24-hour L eq or L dn ). Second, the relationships within the model are expressed as probabilistic rather than deterministic equations. The model has been tested by using acoustical and social survey data collected at 57 sites in the Toronto region exposed to aircraft, road traffic or train noise. Logit analysis was used to estimate two sets of equations. The first predicts the probability of activity interference as a function of event noise level. Four types of interference are included: indoor speech, outdoor speech, difficulty getting to sleep and awakening. The second set predicts the probability of annoyance as a function of the combination of activity interferences. From the first set of equations, it was possible to estimate a function for indoor speech interference only. In this case, the maximum event level was the strongest predictor. The lack of significant results for the other types of interference is explained by the limitations of the data. The same function predicts indoor speech interference for all three sources—road, rail and aircraft noise. The results for the second set of equations show strong relationships between activity interference and the probability of annoyance. Again, the parameters of the logit equations are similar for the three sources. A trial application of the model predicts a higher probability of annoyance for aircraft than for road traffic situations with the same 24-hour L eq . This result suggests that the model may account for previously reported source differences in annoyance.

Approach to interior noise control. I - Damped trim panels

The attenuation mechanisms for interior trim panel systems are reviewed, emphasizing the significance of structure-borne transmission through the trim attachments. The significant factors for high-frequency performance include number of attachments per unit area, panel critical frequency, and panel damping. The need for sufficient damping below and maximal damping above the trim panel critical frequency is described. Described are two experimental flight demonstrations that emphasize the role of trim panel damping. One program involves the new interior system design for a large business jet. Utilization of portions of the weight budget in the form of structural damping treatments permitted a +40% weight savings in acoustical materials with no significant increase in cabin noise levels. Significant reductions in noise levels were achieved in the cockpit, galley, and lavatory. A second program involved a commercial twin-engine jet, with a design objective of significant reduction in noise with minimal weight increase. A reduction of 5 dBA and 5 dB in the speech interference level in the cabin average noise was achieved with only a 0.7% increase in the maximum gross takeoff weight. Only simple add-on skin, trim, and bulkhead damping treatments were used.

Effects of spectral alternation upon intelligibility of words and sentences

In the study of a novel speech interference paradigm, it was found that periodic alternation between nonoverlapping filtering conditions—1100 Hz low pass and 1700 Hz high pass—significantly reduced intelligibility. Thus, although repetition accuracy for CID sentences was approximately 98% in either the low‐pass or high‐pass filtering condition alone, subjects averaged only 48% correct when sentences were subjected to spectral alternation at a rate of 2.67 Hz. Interestingly, the intelligibility of spectrally alternated sentences was dramatically improved by the addition of appropriately filtered noise bands. Thus, when low‐pass noise was mixed with segments of high‐pass speech, and high‐pass noise was mixed with segments of low‐pass speech, repetition accuracy increased by 38.8%. A smaller but statistically significant increase was also found for isolated words. The intelligibility loss produced by spectral alternation appears related to other verbal and musical phenomena. The results from various control conditions indicate that the enhancement of intelligibility produced by alternating noise bands involved separate phonemic restorations for each band. [Work supported by NIH and NSF.]