Intelligibility Of Synthetic Speech Research Articles

A large-scale comparison of two voice synthesis techniques on intelligibility, naturalness, preferences, and attitudes toward voices banked by individuals with amyotrophic lateral sclerosis

Amyotrophic lateral sclerosis (ALS) commonly results in the inability to produce natural speech, making speech-generating devices (SGDs) important. Historically, synthetic voices generated by SGDs were neither unique, nor age- or dialect-appropriate, which depersonalized SGD use. Voices generated by SGDs can now be customized via voice banking and should ideally sound uniquely like the individual’s natural speech, be intelligible, and elicit positive reactions from communication partners. This large-scale 2 x 2 mixed between- and within-participants design examined perceptions of 831 adult listeners regarding custom synthetic voices created for two individuals diagnosed with ALS via two synthesis systems in common clinical use (waveform concatenation and statistical parametric synthesis). The study explored relationships among synthesis system, dysarthria severity, synthetic speech intelligibility, naturalness, and preferences, and also provided a preliminary examination of attitudes regarding the custom synthetic voices. Synthetic voices generated via statistical parametric synthesis trained on deep neural networks were more intelligible, natural, and preferred than voices produced via waveform concatenation, and were associated with more positive attitudes. The custom synthetic voice created from moderately dysarthric speech was more intelligible than the voice created from mildly dysarthric speech. Clinical implications and factors that may have contributed to the relative intelligibilities are discussed.

Case study: Experimental study on optimum level of synthetic speech simulating Lombard effect

The Lombard effect is an acoustic effect associated with human speech in noisy environments. By simulating the Lombard effect in synthetic speech, the intelligibil- ity of synthetic speech in noisy environments is improved. However, it is known that human speech characterized by the Lombard effect increases annoyance in addition to intelligibility. Therefore, the influence of the Lombard effect on synthetic speech in a noisy environment was examined in detail, considering not only intelligibility but also other criteria. It was found that the suitability of speech reproduction in- creased upon using the Lombard effect in a noisy environment. By simulating the Lombard effect in synthetic speech, the intelligibility of synthetic speech in noisy environments is improved. However, it is known that human speech characterized by the Lombard effect increases annoyance in addition to intelligibility. Therefore, the influence of the Lombard effect on synthetic speech in a noisy environment was examined in detail, considering not only intelligibility but also other criteria. It was found that the suitability of speech reproduction increased upon using the Lombard effect in a noisy environment.

Two-Band Radial Postfiltering in Cepstral Domain with Application to Speech Synthesis

Postfiltering is a well known technique that helps increasing the quality of coded speech, enhancing speech intelligibility or alleviating the oversmoothing effect of statistical speech processing methods. This letter presents a new formulation of the radial cepstral postfiltering method that enables the application of different postfiltering factors to low and high frequencies. The transition between bands will be smooth and controllable through an adjustable cut-off frequency. The proposed algorithm can be implemented by means of a simple multiplicative matrix of which an analytical expression is derived. The new method provides a flexible framework to tackle issues related to the quality and intelligibility of synthetic speech.

A Survey of Using Vocal Prosody to Convey Emotion in Robot Speech

The use of speech for robots to communicate with their human users has been facilitated by improvements in speech synthesis technology. Now that the intelligibility of synthetic speech has advanced to the point that speech synthesizers are a widely accepted and used technology, what are other aspects of speech synthesis that can be used to improve the quality of human-robot interaction? The communication of emotions through changes in vocal prosody is one way to make synthesized speech sound more natural. This article reviews the use of vocal prosody to convey emotions between humans, the use of vocal prosody by agents and avatars to convey emotions to their human users, and previous work within the human–robot interaction (HRI) community addressing the use of vocal prosody in robot speech. The goals of this article are (1) to highlight the ability and importance of using vocal prosody to convey emotions within robot speech and (2) to identify experimental design issues when using emotional robot speech in user studies.

Implementation of Phonetic Context Variable Length Unit Selection Module for Malay Text to Speech

Problem statement: The main problem with current Malay Text-To-Speech (MTTS) synthesis system is the poor quality of the generated speech sound due to the inability of traditional TTS system to provide multiple choices of unit for generating more accurate synthesized speech. Approach: This study proposes a phonetic context variable length unit selection MTTS system that is capable of providing more natural and accurate unit selection for synthesized speech. It implemented a phonetic context algorithm for unit selection for MTTS. The unit selection method (without phonetic context) may encounter the problem of selecting the speech unit from different sources and affect the quality of concatenation. This study proposes the design of speech corpus and unit selection method according to phonetic context so that it can select a string of continuous phoneme from same source instead of individual phoneme from different sources. This can further reduce the concatenation point and increase the quality of concatenation. The speech corpus was transcribed according to phonetic context to preserve the phonetic information. This method utilizes word base concatenation method. Firstly it will search through the speech corpus for the target word, if the target is found; it will be used for concatenation. If the word does not exist, then it will construct the words from phoneme sequence. Results: This system had been tested with 40 participants in Mean Opinion Score (MOS) listening test with the average rates for naturalness, pronunciation and intelligibility are 3.9, 4.1 and 3.9. Conclusion/Recommendation: Through this study, a very first version of Corpus-based MTTS has been designed; it has improved the naturalness, pronunciation and intelligibility of synthetic speech. But it still has some lacking that need to be perfected such as the prosody module to support the phrasing analysis and intonation of input text to match with the waveform modifier.

Evaluation of synthetic speech quality: A comparative study of several computer-based speech synthesizers

This paper examines some versions of Lithuanian and Russian synthetic speech intelligibility and Lithuanian, Russian, Hungarian and Italian synthetic speech acceptability. The speech of both Russian and Lithuanian speaker is more intelligible than Russian or Lithuanian synthesis. Previous version of Russian synthesis is worse than Lithuanian and improved Russian synthesis (IRS). Study of characteristics of IRS sounds shows two opposite tendencies - according to the general quantity of mistake reduction this version is tending towards the natural speech, but according to the homogeneity of mistakes, it moves away. As the first tendency is clearly dominant, the general resultant in the new version shows a tend to improve.Correlation between intelligibility and acceptability of IRS deals possibility of small progress towards the natural speech. The IRS is more acceptable to subjects than previous version. The old synthesis is viewed as a rather decent instance of a robot's speech, while the IRS - as a poor variant of human speech.Acceptability studies showed natural speech more enjoyed by Hungarian listeners and more critical by Italian. All versions of synthetic speech were judged as less acceptable than natural but after improvement most of listeners changed their mind.

Intelligibility enhancement of synthetic speech heard via telephone in a noisy environment

The intelligibility of synthetic speech, like the intelligibility of natural speech, decreases if it is transmitted via telephone and/or heard in a noisy environment. A natural speaker adapts his speaking style to these hard speaking conditions in order to restore intelligibility. With new applications for speech synthesis, e.g., navigation systems for cars, the need for adapting synthetic speech to hard speaking conditions arises. Here, the results of intelligibility tests for German databased synthetic speech in comparison to natural speech are presented, and first methods of intelligibility enhancement for synthetic speech are given. It is shown that under the conditions mentioned above unadapted synthetic speech has a lower segmental intelligibility for all phonemes or phoneme clusters than natural speech. For that reason the methods applied affected all speech sounds equally. They are based on prevoius studies of natural speech produced under stress conditions like a noise afflicted environment [Summers et al., J. Acoust. Soc. Am. 84, 917–928 (1988)]. It is shown that the intelligibility of synthetic speech is improved significantly by these enhancement methods. The results give evidence to an individual adaptation for phonemes or phoneme clusters in future.

The Effect of Lexical Complexity on Intelligibility

Most intelligibility tests are based on the use of monosyllabictest stimuli. This constraint eliminates the ability to measurethe effects of lexical stress patterns, complex phonotacticorganizations, and morphological complexity on intelligibility.Since these aspects of lexical structure affect speechproduction (e.g., by changing syllable duration), it is likelythat they affect the structure of acoustic-phonetic patterns.Thus, to the extent that text-to-speech systems fail to modifyacoustic-phonetic patterns appropriately in polysyllabic words,intelligibility may suffer. This means that while most standardintelligibility tests may accurately estimate theintelligibility of monosyllabic words, this estimate may notgeneralize as well to predict the intelligibility of words withmore complex lexical structures. The present study was carriedout to measure how words varying in lexical complexity differ inintelligibility. Monosyllabic, bisyllabic, and polysyllabicwords were used varying in morphological complexity(monomorphemic or polymorphemic). Listeners transcribed thesestimuli spoken by two human talkers and two text-to-speechsystems varying in speech quality. The results indicate thatlexical complexity does affect the measured intelligibility ofsynthetic speech and should be manipulated in order toaccurately predict the performance of text-to-speech systemswith unrestricted natural text.

PURR—A method for prosody evaluation and investigation

Since intelligibility of synthetic speech is no longer the main criterion on which to base quality judgements, reliable methods for prosody evaluation become more important. We propose a method called PURR (Prosody Unveiling through Restricted Representation) to evaluate the prosodic component of a synthesis system without the interference of other system components. In PURR, the stimuli are reduced to their prosodic content. The method has proven to be suitable for test designs with naıve listeners. It can be used for comparative studies as well as for diagnostic analyses and is, therefore, a useful tool for basic research on the perception of prosodic phenomena. In this paper we first describe how the best signal manipulation method was determined using perception tests. The appropriateness of the resulting signal is further assessed in a recognition test of syntactic structure. We then report on further validations of the proposed method: different ways of synthetic prosody modelling are evaluated, both in comparison with human prosody and amongst different synthesis systems.

Effects of listener training on intelligibility of Chinese-accented English

Previous research has demonstrated significant improvements in the intelligibility of synthetic speech following listener training [E. Schwab et al., Human Factors 27, 395–408 (1985)]. In the present study, the effect of listener training on second-language speech intelligibility was investigated. Chinese-accented words and sentences were presented to four groups of native English listeners—one training and one control group for each of two speakers. On word trials, listeners were instructed to select the word presented from two minimal-pair alternatives; on sentence trials, listeners were instructed to write down what they heard. Listener responses were scored as percent of intended words correctly identified. The two training groups participated in three 1-h training sessions. Auditory and visual feedback was provided to listeners during the training sessions only. Performance of the training and control groups was compared across pre- and posttraining test sessions. Improvement in minimal-pair word intelligibility following training was found across speakers for the training group, relative to the control group; a similar but smaller improvement in sentence intelligibility was found. Word results will be discussed in terms of improvement in discriminability following training for the twelve phonemic contrasts tested. [Work supported in part by NIH-NIDCD Grant No. 2R44DC02213.]

Effect of Visual Cuing on Synthetic Speech Intelligibility: A Comparison of Native and Nonnative Speakers of English

This study compared the effect of visual cuing on the intelligibility of DECtalk for native and nonnative speakers of English in both ideal listening conditions and in the presence of background noise at a signal to noise (S/N) ratio of + 10dB. Visual cuing improved DECtalk's intelligibility for nonnative speakers more than for native speakers, especially in the background noise condition. Implications of these findings and the need for further research are discussed.

The effect of lexical complexity on segmental intelligibility.

Most intelligibility tests are based on the use of monosyllabic test stimuli. This constraint eliminates the ability to measure the effects of lexical stress patterns, complex phonotactic organizations, and morphological complexity on segmental intelligibility. Since these aspects of lexical structure affect speech production (e.g., changing syllable duration), it is likely that they affect the structure of acoustic‐phonetic patterns. It also seems plausible that listeners make use of this knowledge during segmental perception. Thus, to the extent that text‐to‐speech systems fail to modify acoustic‐phonetic patterns appropriately in polysyllabic words, intelligibility may suffer. This means that while most standard intelligibility tests may accurately estimate segmental intelligibility in monosyllabic words, this estimate may not generalize prediction of segmental intelligibility with more complex lexical forms. The present study was carried out to measure segmental intelligibility in stimuli varying in lexical complexity. Monosyllabic, bisyllabic, and polysyllabic words were used varying in morphological complexity (monomorphemic or polymorphemic). Listeners transcribed these stimuli spoken by two human talkers and two text‐to‐speech systems varying in speech quality. The results indicate that lexical complexity does affect the measured segmental intelligibility of synthetic speech.

Synthetic speech intelligibility: comparison of native and non-native speakers of English

This study sought to determine the intelligibility of DECTalkTMsynthesized speech to both native and non-native speakers of English. The ability of the groups to understand DECTalk sentences delivered both in quiet and at a signal-to-noise (S/N) ratio of +10 dB was investigated. Results suggested that non-native speakers made significantly more errors transcribing DECTalk sentences than did native speakers. DECTalk was found to be significantly less intelligible in noise than in quiet for both groups, but non-native speakers experienced significantly more difficulty understanding DECTalk in noise than did native speakers. Implications of these findings and the need for further research are discussed.

A formant bandwidth estimation procedure for vowel synthesis [43.72.Ja

The specification of vowel formant bandwidths for speech synthesis has been inconsistent in the past, perhaps due to the difficulty of measuring formant bandwidths in natural speech and the possible perceptual insignificance of formant bandwidths on the intelligibility of synthetic speech. Here, regression equations are presented for the estimation of formant bandwidths based on measurements from natural speech which is based only on formant center frequency and independent of other formant values. Current usage, as well as comparison with another well-known estimation algorithm suggests that the new procedure should be quite acceptable for some types of speech synthesis.

Synthetic speech intelligibility under several experimental conditions

Present technology makes it possible for nonspeaking individuals to communicate using electronic augmentative communication devices that generate synthetic speech. For augmented individuals to communicate effectively, the synthetic speech must be intelligible, not just in ideal listening conditions, but in the presence of background noise. The present study sought to determine the extent to which two of the most frequently researched DECTalk™ voices, DECPaul and DECBetty, degraded in the presence of background noise at a signal-tonoise (S/N) ratio of +10 dB. It further sought to determine the effect of adding visual cuing to synthetic speech output on the intelligibility of DECTalk. The results suggested that DECTalk is significantly less intelligible in the presence of background noise at a +10 dB S/N ratio than in ideal listening conditions. Although the addition of visual cuing did not increase intelligibility overall, subjects did have fewer sentence errors when visual cuing was added to synthetic spe...

Intelligibility of Synthetic Speech for Normal-Hearing and Hearing-Impaired Listeners

This study examined the perceived intelligibility of synthetic speech. Participants were adults aged 49-69, one group with normal hearing and one group with acquired sensorineural hearing impairment. Word lists were presented in two speech types: DECtalk (a high-quality speech synthesizer) and a natural male speaker. Results revealed differences between groups, with normal-hearing listeners scoring higher than hearing-impaired listeners, and between speech types, with higher scores for natural speech than for synthesized speech. There was no significant interaction of hearing level and speech type.

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.

Segmental intelligibility of synthetic speech produced by rule

This paper reports the results of an investigation that employed the modified rhyme test (MRT) to measure the segmental intelligibility of synthetic speech generated automatically by rule. Synthetic speech produced by ten text-to-speech systems was studied and compared to natural speech. A variation of the standard MRT was also used to study the effects of response set size on perceptual confusions. Results indicated that the segmental intelligibility scores formed a continuum. Several systems displayed very high levels of performance that were close to or equal to scores obtained with natural speech; other systems displayed substantially worse performance compared to natural speech. The overall performance of the best system, DECtalk--Paul, was equivalent to the data obtained with natural speech for consonants in syllable-initial position. The findings from this study are discussed in terms of the use of a set of standardized procedures for measuring intelligibility of synthetic speech under controlled laboratory conditions. Recent work investigating the perception of synthetic speech under more severe conditions in which greater demands are made on the listener's processing resources is also considered. The wide range of intelligibility scores obtained in the present study demonstrates important differences in perception and suggests that not all synthetic speech is perceptually equivalent to the listener.

The Intelligibility of Synthetic Speech to Learning Handicapped Children

This study investigated the effect of listening experience on the ability of learning handicapped children to understand synthetic speech. Two groups of 12 learning handicapped children each participated in the study. The first group was exposed to synthetic speech every day for 10 weeks, whereas the second had limited or no exposure to synthetic speech. When children were asked to identify stimulus words presented by a taped, female voice, the group with extended exposure to synthetic speech correctly identified approximately the same percentage of words as the group with limited exposure In contrast, when the stimulus words were presented by a speech synthesizer, children in the extended exposure group correctly identified a significantly higher percentage of the words than did those in the limited exposure group. Both groups correctly identified significantly fewer wends when words were presented by the speech synthesizer than when words were presented by the tape recorder. Educational implications of the low intelligibility scores of both groups when listening to synthesized speech are discussed, and future research needs identified.

Context, Repetition and Synthesized Speech Intelligibility

Two studies were conducted to explore the effects of various parameters on rule-based synthetic speech intelligibility. Experiment I examined the effect of situational context clues and speech rate on synthesized speech intelligibility. Subjects who received pragmatic context information prior to each message had transcription error rates 50% lower than those who received no context information. Speech rates of 250 words per minute (wpm) yielded significantly more transcription errors than rates of 180 wpm. In Experiment II, the effects of speech rate, message repetition, and location of information in a message were examined. Transcription accuracy was best for messages spoken at 150 or 180 wpm and for messages repeated either twice or three times. Words at the end of messages were transcribed more accurately than words at the beginning of messages. Subjective ratings indicated that subjects were aware of errors when incorrectly transcribing a message even though no feedback was provided.