Intelligibility Of Synthesized Speech Research Articles

Subjective and objective measurement of synthesized speech intelligibility in modern telephone conditions

This paper investigates the impact of different telephone channels, represented by impairments as introduced by modern telecommunication networks (e.g. speech coding, bandwidth limitation, packet loss, etc.), on the intelligibility of synthesized speech. Both subjective and objective assessments are used. Two different speech intelligibility prediction models, namely PESQ Intelligibility and POLQA Intelligibility, are evaluated by comparing the predictions with subjectively obtained intelligibility scores. The results show that all the investigated degradations seriously impact the intelligibility of the synthesized speech measured subjectively. Furthermore it is shown that PESQ Intelligibility provides too low correlations between predicted objective measurements and subjective scores for accurate prediction of speech intelligibility while POLQA Intelligibility is capable of providing good intelligibility predictions in the case that a closed response experimental set up is used.

Synthesized Speech Intelligibility and Early Preschool-Age Children: Comparing Accuracy for Single-Word Repetition With Repeated Exposure

This investigation examined the effect of repeated exposure to novel and repeated spoken words in typical environments on the intelligibility of 2 synthesized voices and human recorded speech in preschools. Eighteen preschoolers listened to and repeated single words presented in human-recorded speech, DECtalk Paul, and AT&T Voice Michael during 5 experimental sessions. Stimuli consisted of repeated and novel words presented in each speech output condition during each session. Sessions took place in the presence of typically occurring noise in classroom or home settings. There was a significant main effect for voice as participants accurately identified significantly more words in the human-recorded speech and AT&T Voice than in the DECtalk speech output condition. When averaged across speech output conditions, children increased their accuracy as they participated in additional sessions. There was a statistically significant interaction between session and voice. DECtalk had a slightly larger effect of session than did AT&T Voice and human-recorded speech.

Synthesized Speech Output and Children: A Scoping Review

Many computer-based augmentative and alternative communication systems in use by children have speech output. This article (a) provides a scoping review of the literature addressing the intelligibility and listener comprehension of synthesized speech output with children and (b) discusses future research directions. Studies investigating synthesized speech intelligibility and/or comprehension with children as listeners were systematically identified and coded according to their objectives and methodology. Ten studies were identified. They were organized according to the following variables: intelligibility variables related to the stimuli (context and rate), intelligibility variables related to aspects of the listener (age of the child, the language or languages spoken by the listener, experience, and practice effects), and comprehension. Each of these factors-and the research support with child participants-was discussed. Overall, there is a paucity of research investigating synthesized speech for use with children. Available evidence suggests that children produce similar trends but lower levels of intelligibility performance when compared with adults. Future areas of applied research are required to adequately define this relationship and the variables that may contribute to improving the intelligibility and comprehension of synthesized speech for children.

Synthesized speech intelligibility and persuasion: Speech rate and non-native listeners

This experiment assessed the effect of variation in speech rate on comprehension and persuasiveness of a message presented in text-to-speech (TTS) synthesis to native and non-native listeners. Eighty non-native speakers of English and 80 native speakers of Australian English were randomly assigned to listen to eight banking product descriptions under one of four conditions: normal rate (155 words per minute) with no background noise, normal rate with multi-talker background noise (+6 dB SNR), fast-normal (178 words per minute) with no background noise, and fast-normal with multi-talker background noise. Participants completed comprehension tests and rated each product’s usefulness. A faster rate lowers comprehension for both native and non-native listeners but does not influence the persuasiveness of the message. The findings have implications for the selection of speech rates for persuasive messages delivered to native and non-native listeners using TTS.

Synthesized speech intelligibility among native speakers and non-native speakers of English

Using synthesized and digitized speech in electronic communication devices may greatly benefit individuals who cannot produce intelligible speech. However, multiple investigations have demonstrated that synthesized speech is not always sufficiently intelligible for its listeners. Listening to synthesized speech may be particularly problematic for listeners for whom English is a second language. We compared native and non-native English-speaking adults' listening accuracy for English sentences in natural voice and synthesized voice conditions. Results indicated a disproportionate disadvantage for the non-native English-speaking group when listening to synthesized speech compared to their native English-speaking age peers. There was, however, significant variability in performance within the non-native English group, and this was strongly related to independent measures of English language skill. Specifically, a large portion of the variance in performance on the synthesized speech task was predicted by participants' receptive vocabulary scores.

Synthesized Speech Intelligibility in Sentences

Research comparing the intelligibility of human and synthesized speech among both young children and adults has indicated that synthesized speech results in a degrading of intelligibility. The purpose of this study was to compare speech intelligibility of high-probability sentences produced using DECtalk Perfect Paul and live speech among monolingual English-speaking and bilingual children. Twenty typically developing children between the ages of 4;5 (years;months) and 6;2 participated. Ten children (6 boys and 4 girls) were monolingual English speakers. The remaining ten children (4 boys and 6 girls) were bilingual. Their reproduction of modeled sentences produced in English via live speech and synthesized speech were analyzed to determine the intelligibility of each speech type. The results indicated that for both groups of children, performance was significantly better in the live speech condition. Results further revealed that the intelligibility decrement for synthesized speech was substantially greater for bilingual speakers. For bilingual children, even higher quality synthesized speech may present a greater challenge than it does for monolingual English-speaking children. Possible strategies to enhance intelligibility are discussed.

The intelligibility of synthesized speech in data inquiry systems.

Recent improvements in speech technology have made synthetic speech a viable form of system feedback. However, little is known about the effects of various speech parameters on speech intelligibility and system performance. Experiment 1 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 and longer response latencies than rates of 180 wpm. In Experiment 2 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.

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.