skip to main content


Title: Perception of Concatenative vs. Neural Text-To-Speech (TTS): Differences in Intelligibility in Noise and Language Attitudes
This study tests speech-in-noise perception and social ratings of speech produced by different text-to-speech (TTS) synthesis methods. We used identical speaker training datasets for a set of 4 voices (using AWS Polly TTS), generated using neural and concatenative TTS. In Experiment 1, listeners identified target words in semantically predictable and unpredictable sentences in concatenative and neural TTS at two noise levels (-3 dB, -6 dB SNR). Correct word identification was lower for neural TTS than for concatenative TTS, in the lower SNR, and for semantically unpredictable sentences. In Experiment 2, listeners rated the voices on 4 social attributes. Neural TTS was rated as more human-like, natural, likeable, and familiar than concatenative TTS. Furthermore, how natural listeners rated the neural TTS voice was positively related to their speech-in-noise accuracy. Together, these findings show that the TTS method influences both intelligibility and social judgments of speech — and that these patterns are linked. Overall, this work contributes to our understanding of the of the nexus of speech technology and human speech perception.  more » « less
Award ID(s):
1911855
NSF-PAR ID:
10275344
Author(s) / Creator(s):
;
Date Published:
Journal Name:
Proceedings of Interspeech
Page Range / eLocation ID:
1733 to 1737
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Building on previous work in subset selection of training data for text-to-speech (TTS), this work compares speaker-level and utterance-level selection of TTS training data, using acoustic features to guide selection. We find that speaker-based selection is more effective than utterance-based selection, regardless of whether selection is guided by a single feature or a combination of features. We use US English telephone data collected for automatic speech recognition to simulate the conditions of TTS training on low-resource languages. Our best voice achieves a human-evaluated WER of 29.0% on semantically-unpredictable sentences. This constitutes a significant improvement over our baseline voice trained on the same amount of randomly selected utterances, which performed at 42.4% WER. In addition to subjective voice evaluations with Amazon Mechanical Turk, we also explored objective voice evaluation using mel-cepstral distortion. We found that this measure correlates strongly with human evaluations of intelligibility, indicating that it may be a useful method to evaluate or pre-select voices in future work. 
    more » « less
  2. null (Ed.)
    The present study compares how individuals perceive gradient acoustic realizations of emotion produced by a human voice versus an Amazon Alexa text-to-speech (TTS) voice. We manipulated semantically neutral sentences spoken by both talkers with identical emotional synthesis methods, using three levels of increasing ‘happiness’ (0 %, 33 %, 66% ‘happier’). On each trial, listeners (native speakers of American English, n=99) rated a given sentence on two scales to assess dimensions of emotion: valence (negative-positive) and arousal (calm-excited). Participants also rated the Alexa voice on several parameters to assess anthropomorphism (e.g., naturalness, human-likeness, etc.). Results showed that the emotion manipulations led to increases in perceived positive valence and excitement. Yet, the effect differed by interlocutor: increasing ‘happiness’ manipulations led to larger changes for the human voice than the Alexa voice. Additionally, we observed individual differences in perceived valence/arousal based on participants’ anthropomorphism scores. Overall, this line of research can speak to theories of computer personification and elucidate our changng relationship with voice-AI technology. 
    more » « less
  3. Listening to speech in noise can require substantial mental effort, even among younger normal-hearing adults. The task-evoked pupil response (TEPR) has been shown to track the increased effort exerted to recognize words or sentences in increasing noise. However, few studies have examined the trajectory of listening effort across longer, more natural, stretches of speech, or the extent to which expectations about upcoming listening difficulty modulate the TEPR. Seventeen younger normal-hearing adults listened to 60-s-long audiobook passages, repeated three times in a row, at two different signal-to-noise ratios (SNRs) while pupil size was recorded. There was a significant interaction between SNR, repetition, and baseline pupil size on sustained listening effort. At lower baseline pupil sizes, potentially reflecting lower attention mobilization, TEPRs were more sustained in the harder SNR condition, particularly when attention mobilization remained low by the third presentation. At intermediate baseline pupil sizes, differences between conditions were largely absent, suggesting these listeners had optimally mobilized their attention for both SNRs. Lastly, at higher baseline pupil sizes, potentially reflecting over-mobilization of attention, the effect of SNR was initially reversed for the second and third presentations: participants initially appeared to disengage in the harder SNR condition, resulting in reduced TEPRs that recovered in the second half of the story. Together, these findings suggest that the unfolding of listening effort over time depends critically on the extent to which individuals have successfully mobilized their attention in anticipation of difficult listening conditions. 
    more » « less
  4. The meaning of words in natural language depends crucially on context. However, most neuroimaging studies of word meaning use isolated words and isolated sentences with little context. Because the brain may process natural language differently from how it processes simplified stimuli, there is a pressing need to determine whether prior results on word meaning generalize to natural language. fMRI was used to record human brain activity while four subjects (two female) read words in four conditions that vary in context: narratives, isolated sentences, blocks of semantically similar words, and isolated words. We then compared the signal-to-noise ratio (SNR) of evoked brain responses, and we used a voxelwise encoding modeling approach to compare the representation of semantic information across the four conditions. We find four consistent effects of varying context. First, stimuli with more context evoke brain responses with higher SNR across bilateral visual, temporal, parietal, and prefrontal cortices compared with stimuli with little context. Second, increasing context increases the representation of semantic information across bilateral temporal, parietal, and prefrontal cortices at the group level. In individual subjects, only natural language stimuli consistently evoke widespread representation of semantic information. Third, context affects voxel semantic tuning. Finally, models estimated using stimuli with little context do not generalize well to natural language. These results show that context has large effects on the quality of neuroimaging data and on the representation of meaning in the brain. Thus, neuroimaging studies that use stimuli with little context may not generalize well to the natural regime.

    SIGNIFICANCE STATEMENTContext is an important part of understanding the meaning of natural language, but most neuroimaging studies of meaning use isolated words and isolated sentences with little context. Here, we examined whether the results of neuroimaging studies that use out-of-context stimuli generalize to natural language. We find that increasing context improves the quality of neuro-imaging data and changes where and how semantic information is represented in the brain. These results suggest that findings from studies using out-of-context stimuli may not generalize to natural language used in daily life.

     
    more » « less
  5. This paper describes experiments in training HMM-based text-to-speech (TTS) voices on data collected for Automatic Speech Recognition (ASR) training. We compare a number of filtering techniques designed to identify the best utterances from a noisy, multi-speaker corpus for training voices, to exclude speech containing noise and to include speech close in nature to more traditionally-collected TTS corpora. We also evaluate the use of automatic speech recognizers for intelligibility assessment in comparison with crowdsourcing methods. While the goal of this work is to develop natural-sounding and intelligible TTS voices in Low Resource Languages (LRLs) rapidly and easily, without the expense of recording data specifically for this purpose, we focus on English initially to identify the best filtering techniques and evaluation methods. We find that, when a large amount of data is available, selecting from the corpus based on criteria such as standard deviation of f0, fast speaking rate, and hypo-articulation produces the most intelligible voices. 
    more » « less