More Like this

1. Characteristics of Text-to-Speech and Other Corpora

   Cooper, Erica ; Li, Emily ; Hirschberg, Julia ( June 2018 , Proceedings of Speech Prosody 2018)

   Extensive TTS corpora exist for commercial systems cre- ated for high-resource languages such as Mandarin, English, and Japanese. Speakers recorded for these corpora are typically instructed to maintain constant f0, energy, and speaking rate and are recorded in ideal acoustic environments, producing clean, consistent audio. We have been developing TTS systems from “found” data collected for other purposes (e.g. training ASR systems) or available on the web (e.g. news broadcasts, au- diobooks) to produce TTS systems for low-resource languages (LRLs) which do not currently have expensive, commercial sys- tems. This study investigates whether traditional TTS speakers do exhibit significantly less variation and better speaking char- acteristics than speakers in found genres. By examining char- acteristics of f0, energy, speaking rate, articulation, NHR, jit- ter, and shimmer in found genres and comparing these to tra- ditional TTS corpora, We have found that TTS recordings are indeed characterized by low mean pitch, standard deviation of energy, speaking rate, and level of articulation, and low mean and standard deviations of shimmer and NHR; in a number of respects these are quite similar to some found genres. By iden- tifying similarities and differences, we are able to identify ob- jective methods for selecting found data to build TTS systems for LRLs. 

   more » « less
2. Utterance Selection for Optimizing Intelligibility of TTS Voices Trained on ASR Data

   https://doi.org/DOI: 10.21437/Interspeech.2017-465  

   Erica Cooper, Xinyue Wang ( January 2017 , Interspeech 2017)

   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
3. Subset Selection, Adaptation, Gemination and Prosody Prediction for Amharic Text-to-Speech Synthesis

   https://doi.org/10.21437/SSW.2019-37  

   Tesfaye Biru, Elshadai ; Tofik Mohammed, Yishak ; Tofu, David ; Cooper, Erica ; Hirschberg, Julia ( September 2019 , 10th ISCA Speech Synthesis Workshop)

   While large TTS corpora exist for commercial sys- tems created for high-resource languages such as Man- darin, English, and Spanish, for many languages such as Amharic, which are spoken by millions of people, this is not the case. We are working with “found” data collected for other purposes (e.g. training ASR systems) or avail- able on the web (e.g. news broadcasts, audiobooks) to produce TTS systems for low-resource languages which do not currently have expensive, commercial systems. This study describes TTS systems built for Amharic from “found” data and includes systems built from di erent acoustic-prosodic subsets of the data, systems built from combined high and lower quality data using adaptation, and systems which use prediction of Amharic gemination to improve naturalness as perceived by evaluators. 

   more » « less
4. Individual Variation in Language Attitudes Toward Voice-AI: The Role of Listeners’ Autistic-Like Traits

   https://doi.org/10.21437/Interspeech.2020-1339  

   Cohn, Michelle ; Sarian, Melina ; Predeck, Kristin ; Zellou, Georgia ( October 2020 , Proceedings of Interspeech)

   null (Ed.)

   More and more, humans are engaging with voice-activated artificially intelligent (voice-AI) systems that have names (e.g., Alexa), apparent genders, and even emotional expression; they are in many ways a growing ‘social’ presence. But to what extent do people display sociolinguistic attitudes, developed from human-human interaction, toward these disembodied text-to-speech (TTS) voices? And how might they vary based on the cognitive traits of the individual user? The current study addresses these questions, testing native English speakers’ judgments for 6 traits (intelligent, likeable, attractive, professional, human-like, and age) for a naturally-produced female human voice and the US-English default Amazon Alexa voice. Following exposure to the voices, participants completed these ratings for each speaker, as well as the Autism Quotient (AQ) survey, to assess individual differences in cognitive processing style. Results show differences in individuals’ ratings of the likeability and human-likeness of the human and AI talkers based on AQ score. Results suggest that humans transfer social assessment of human voices to voice-AI, but that the way they do so is mediated by their own cognitive characteristics. 

   more » « less
5. Quantitative Acoustic versus Deep Learning Metrics of Lenition

   https://doi.org/10.3390/languages8020098  

   Wayland, Ratree ; Tang, Kevin ; Wang, Fenqi ; Vellozzi, Sophia ; Sengupta, Rahul ( June 2023 , Languages)

   Spanish voiced stops /b, d, ɡ/ surfaced as fricatives [β, ð, ɣ] in intervocalic position due to a phonological process known as spirantization or, more broadly, lenition. However, conditioned by various factors such as stress, place of articulation, flanking vowel quality, and speaking rate, phonetic studies reveal a great deal of variation and gradience of these surface forms, ranging from fricative-like to approximant-like [β⊤, ð⊤, ɣ⊤]. Several acoustic measurements have been used to quantify the degree of lenition, but none is standard. In this study, the posterior probabilities of sonorant and continuant phonological features in a corpus of Argentinian Spanish estimated by a deep learning Phonet model as measures of lenition were compared to traditional acoustic measurements of intensity, duration, and periodicity. When evaluated against known lenition factors: stress, place of articulation, surrounding vowel quality, word status, and speaking rate, the results show that sonorant and continuant posterior probabilities predict lenition patterns that are similar to those predicted by relative acoustic intensity measures and are in the direction expected by the effort-based view of lenition and previous findings. These results suggest that Phonet is a reliable alternative or additional approach to investigate the degree of lenition. 

   more » « less