{"title":"Text Normalization for Speech Systems for All Languages","authors":"Athiya Deviyani, A. Black","doi":"10.21437/s4sg.2022-5","DOIUrl":"https://doi.org/10.21437/s4sg.2022-5","url":null,"abstract":"Most text-to-speech systems suffer from the limitation that its inputs should be a set of strings of characters with standard pronunciation, and struggle when given input is in the form of symbols, numbers, or abbreviations that often occur in real text. One of the most common ways to address this problem is to automatically map non-standard words to standard words using statistical, neural and rule-driven methods. However, despite the significant efforts of normalizing such words, there is just too much variability in existing corpora such that it is extremely challenging to capture edge cases. In this work, we propose a tool which aids data collection from (non-programmer) native speakers to allow numbers and other common non-standard words to be mapped to standard words that can be pronounced correctly by a synthesizer, while addressing related problems such as identify-ing common non-standard words appear in text and how do we ask questions from native speakers to get sufficient information to allow a useful normalization of non-standard words.","PeriodicalId":359058,"journal":{"name":"1st Workshop on Speech for Social Good (S4SG)","volume":"126 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133407559","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Cross-Teager Cepstral Coefficients For Dysarthric Severity Level Classification","authors":"Anand Therattil, Aastha Kachhi, H. Patil","doi":"10.21437/s4sg.2022-7","DOIUrl":"https://doi.org/10.21437/s4sg.2022-7","url":null,"abstract":"","PeriodicalId":359058,"journal":{"name":"1st Workshop on Speech for Social Good (S4SG)","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114487872","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Yixuan Zhang, Yuanyuan Zhang, T. Patel, O. Scharenborg
{"title":"Comparing data augmentation and training techniques to reduce bias against non-native accents in hybrid speech recognition systems","authors":"Yixuan Zhang, Yuanyuan Zhang, T. Patel, O. Scharenborg","doi":"10.21437/s4sg.2022-4","DOIUrl":"https://doi.org/10.21437/s4sg.2022-4","url":null,"abstract":"One important problem that needs tackling for wide deployment of Automatic Speech Recognition (ASR) is the bias in ASR, i.e., ASRs tend to generate more accurate predictions for certain speaker groups while making more errors on speech from other groups. We aim to reduce bias against non-native speakers of Dutch compared to native Dutch speakers. We investigate three different data augmentation techniques - speed and volume perturbation and pitch shift - to increase the amount of non-native accented Dutch training data, and use the augmented data for two transfer learning techniques: model fine-tuning and multi-task learning, to reduce bias in a state-of-the-art hybrid HMM-DNN Kaldi-based ASR system. Experimental results on Dutch read speech and human-machine interaction (HMI) speech showed that although individual data augmentation techniques did not always yield an improved recognition performance, the combination of all three did. Importantly, bias was reduced by more than 18% absolute compared to the base-line system for read speech when applying pitch shift and multi-task training, and by more than 7% for HMI speech when applying all three data augmentation techniques during fine-tuning, while improving recognition accuracy of both native and non-native Dutch speech.","PeriodicalId":359058,"journal":{"name":"1st Workshop on Speech for Social Good (S4SG)","volume":"31 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127958253","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Towards an Automatic Speech Recognizer for the Choctaw language","authors":"Jacqueline Brixey, D. Traum","doi":"10.21437/s4sg.2022-2","DOIUrl":"https://doi.org/10.21437/s4sg.2022-2","url":null,"abstract":"In this work we describe ongoing development of the first automatic speech recognition (ASR) system for the American indigenous language, Choctaw (ISO 639-2: cho, endonym: Chahta). Choctaw is spoken by the Choctaw people, with an estimated 10,000 fluent speakers across three federally recognized Choctaw tribes. The Choctaw language is subject-objectverb order, and is highly inflectional, with prefixes, suffixes, and infixes possible on a single verb base. The language also has rhythmic lengthening, in which certain vowels are lengthened based on vowels in affixes. The motivation for developing an ASR system include: assisting in efforts to revitalize and reclaim the endangered language by aiding language learners; promoting additional contexts and scenarios for increased language use, such as conversations with automated dialogue systems; and supporting language documentation. We describe our collection of two-party conversational data and repetition of prepared phrases from a diverse set of speakers that was used to train the system. The ASR model was implemented using Kaldi. The model is currently trained and tested on a subset of the collected data, and achieves a WER of 49.35%.","PeriodicalId":359058,"journal":{"name":"1st Workshop on Speech for Social Good (S4SG)","volume":"41 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129723656","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Takuma Yoshimoto, R. Takashima, Chiho Sasaki, T. Takiguchi
{"title":"Highly Intelligible Speech Synthesis for Spinal Muscular Atrophy Patients Based on Model Adaptation","authors":"Takuma Yoshimoto, R. Takashima, Chiho Sasaki, T. Takiguchi","doi":"10.21437/s4sg.2022-8","DOIUrl":"https://doi.org/10.21437/s4sg.2022-8","url":null,"abstract":"Recently, speech signal processing technology has been used to assist people with disabilities, and the demand for such technology is increasing. In this study, we focus on spinal muscular atrophy (SMA) patients. SMA is a neuromuscular disease. Those with this disease have speech that is unclear compared to that of normal subjects which results from the use of a ventilator af-ter a tracheotomy and from the atrophy of muscles that move the mouth, etc. Therefore, it is difficult to understand what they are saying, making communication difficult. In this paper, we analyze the speech of people with SMA and propose a text-to-speech (TTS) system to aid in communication. The proposed system uses an approach that adapts a TTS model pre-trained using normal speech to speech of a person with SMA. This system can synthesize speech having both of intelligibility derived from normal speech and individuality derived from the speech of the target subject with SMA.","PeriodicalId":359058,"journal":{"name":"1st Workshop on Speech for Social Good (S4SG)","volume":"31 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123916273","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Satwik Dutta, Jacob C. Reyna, J. Buzhardt, Dwight W. Irvin, John H. L. Hansen
{"title":"Can Smartphones be a cost-effective alternative to LENA for Early Childhood Language Intervention?","authors":"Satwik Dutta, Jacob C. Reyna, J. Buzhardt, Dwight W. Irvin, John H. L. Hansen","doi":"10.21437/s4sg.2022-3","DOIUrl":"https://doi.org/10.21437/s4sg.2022-3","url":null,"abstract":"Although non-profit commercial products such as LENA can provide valuable feedback to parents and early childhood edu-cators about their children’s or student’s daily communication interactions, their cost and technology requirements put them out of reach of many families who could benefit. Over the last two decades, smartphones have become commonly used in most households irrespective of their socio-economic background. In this study, conducted during the COVID-19 pandemic, we aim to compare audio collected on LENA recorders versus smartphones available to families in an unsupervised data collection protocol. Approximately 10 hours of audio evaluated in this study was collected by three families in their homes during parent-child science book reading activities with their children. We report comparisons and found similar performance between the two audio capture devices based on their speech signal-to-noise ratio (NIST STNR) and word-error-rates calculated using automatic speech recognition (ASR) engines. Finally, we discuss implications of this study for expanding this technology to more diverse populations, limitations and future directions.","PeriodicalId":359058,"journal":{"name":"1st Workshop on Speech for Social Good (S4SG)","volume":"152 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115793449","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}