We present free high quality multi-speaker speech corpora for Gujarati, Kannada, Malayalam, Marathi, Tamil and Telugu, which are six of the twenty two official languages of India spoken by 374 million native speakers. The datasets are primarily intended for use in text-to-speech (TTS) applications, such as constructing multilingual voices or being used for speaker or language adaptation. Most of the corpora (apart from Marathi, which is a female-only database) consist of at least 2,000 recorded lines from female and male native speakers of the language. We present the methodological details behind corpora acquisition, which can be scaled to acquiring data for other languages of interest. We describe the experiments in building a multilingual text-to-speech model that is constructed by combining our corpora. Our results indicate that using these corpora results in good quality voices, with Mean Opinion Scores (MOS) > 3.6, for all the languages tested. We believe that these resources, released with an open-source license, and the described methodology will help in the progress of speech applications for the languages described and aid corpora development for other, smaller, languages of India and beyond.

This paper introduces an open-source crowd-sourced multi-speaker speech corpus along with the comprehensive set of finite-state transducer (FST) grammars for performing text normalization for the Burmese (Myanmar) language. We also introduce the open-source finite-state grammars for performing grapheme-to-phoneme (G2P) conversion for Burmese. These three components are necessary (but not sufficient) for building a high-quality text-to-speech (TTS) system for Burmese, a tonal Southeast Asian language from the Sino-Tibetan family which presents several linguistic challenges. We describe the corpus acquisition process and provide the details of our finite state-based approach to Burmese text normalization and G2P. Our experiments involve building a multi-speaker TTS system based on long short term memory (LSTM) recurrent neural network (RNN) models, which were previously shown to perform well for other languages in a low-resource setting. Our results indicate that the data and grammars that we are announcing are sufficient to build reasonably high-quality models comparable to other systems. We hope these resources will facilitate speech and language research on the Burmese language, which is considered by many to be low-resource due to the limited availability of free linguistic data.

Stochastic finite automata arise naturally in many language and speech processing tasks. They include stochastic acceptors, which represent certain probability distributions over random strings. We consider the problem of efficient sampling: drawing random string variates from the probability distribution represented by stochastic automata and transformations of those. We show that path-sampling is effective and can be efficient if the epsilon-graph of a finite automaton is acyclic. We provide an algorithm that ensures this by conflating epsilon-cycles within strongly connected components. Sampling is also effective in the presence of non-injective transformations of strings. We illustrate this in the context of decoding for Connectionist Temporal Classification (CTC), where the predictive probabilities yield auxiliary sequences which are transformed into shorter labeling strings. We can sample efficiently from the transformed labeling distribution and use this in two different strategies for finding the most probable CTC labeling.

This paper describes the development of text-to-speech corpora \nfor four South African languages. The approach followed investigated \nthe possibility of using low-cost methods including informal \nrecording environments and untrained volunteer speakers. \nThis objective and the additional future goal of expanding \nthe corpus to increase coverage of South Africa’s 11 official \nlanguages necessitated experimenting with multi-speaker and \ncode-switched data. The process and relevant observations are \ndetailed throughout. The latest version of the corpora are available \nfor download under an open-source license and will likely \nsee further development and refinement in future. \nIndex Terms: text-to-speech corpus, under-resourced languages

We present a large scale effort to build a commercial Automatic Speech Recognition (ASR) product for Arabic. Our goal is to support voice search, dictation, and voice control for the general Arabic-speaking public, including support for multiple Arabic dialects. We describe our ASR system design and compare recognizers for five Arabic dialects, with the potential to reach more than 125 million people in Egypt, Jordan, Lebanon, Saudi Arabia, and the United Arab Emirates (UAE). We compare systems built on diacritized vs. non-diacritized text. We also conduct cross-dialect experiments, where we train on one dialect and test on the others. Our average word error rate (WER) is 24.8% for voice search.

We describe our efforts in deploying Google search by voice for Cantonese, a southern Chinese dialect widely spoken in and around Hong Kong and Guangzhou. We collected audio data from local Cantonese speakers in Hong Kong and Guangzhou by using our DataHound smartphone application. This data was used to create appropriate acoustic models. Language models were trained on anonymized query logs from Google Web Search for Hong Kong. Because users in Hong Kong frequently mix English and Cantonese in their queries, we designed our system from the ground up to handle both languages. We report on experiments with different techniques for mapping the phoneme inventories for both languages into a common space. Based on extensive experiments we report word error rates and web scores for both Hong Kong and Guangzhou data. Cantonese Google search by voice was launched in December 2010. Index Terms: voice search, Cantonese speech recognition, multilingual speech recognition

In prior work, adults with intellectual disabilities answered comprehension questions after reading texts. We apply a latent trait model to this data to infer the intrinsic difficulty of texts for the participant group. We then analyze the correlation between grade levels predicted by an automatic readability assessment tool and the inferred text difficulty.

In this paper we describe our efforts to build a Mandarin Chinese voice search system. We describe our strategies for data collection, language, lexicon and acoustic modeling, as well as issues related to text normalization that are an integral part of building voice search systems. We show excellent performance on typical spoken search queries under a variety of accents and acoustic conditions. The system has been in operation since October 2009 and has received very positive user reviews.

Pronunciation information is available in large quantities on the Web, in the form of IPA and ad-hoc transcriptions. We describe techniques for extracting candidate pronunciations from Web pages and associating them with orthographic words, filtering out poorly extracted pronunciations, normalizing IPA pronunciations to better conform to a common transcription standard, and generating phonemic from ad-hoc transcriptions. We show improvements on a letter-to-phoneme task when using web-derived vs. Pronlex pronunciations.

Adding punctuation and capitalization greatly improves the readability of automatic speech transcripts. We discuss an approach for performing both tasks in a single pass using a purely text-based n-gram language model. We study the effect on performance of varying the n-gram order (from n = 3 to n = 6) and the amount of training data (from 58 million to 55 billion tokens). Our results show that using larger training data sets consistently improves performance, while increasing the n-gram order does not help nearly as much.

Indexing and retrieval of speech content in various forms such as broadcast news, customer care data and on-line media has gained a lot of interest for a wide range of applications, from customer analytics to on-line media search. For most retrieval applications, the speech content is typically first converted to a lexical or phonetic representation using automatic speech recognition (ASR). The first step in searching through indexes built on these representations is the generation of pronunciations for named entities and foreign language query terms. This paper summarizes the results of the work conducted during the 2008 JHU Summer Workshop by the Multilingual Spoken Term Detection team, on mining the web for pronunciations and analyzing their impact on spoken term detection. We will first present methods to use the vast amount of pronunciation information available on the Web, in the form of IPA and ad-hoc transcriptions. We describe techniques for extracting candidate pronunciations from Web pages and associating them with orthographic words, filtering out poorly extracted pronunciations, normalizing IPA pronunciations to better conform to a common transcription standard, and generating phonemic representations from ad-hoc transcriptions. We then present an analysis of the effectiveness of using these pronunciations to represent Out-Of-Vocabulary (OOV) query terms on the performance of a spoken term detection (STD) system. We will provide comparisons of Web pronunciations against automated techniques for pronunciation generation as well as pronunciations generated by human experts. Our results cover a range of speech indexes based on lattices, confusion networks and one-best transcriptions at both word and word fragments levels.

Censored targets, such as the time to events in survival analysis, can generally be represented by intervals on the real line. In this paper, we propose a novel support vector technique (named SVCR) for regression on censored targets. SVCR inherits the strengths of support vector methods, such as a globally optimal solution by convex programming, fast training speed and strong generalization capacity. In contrast to ranking approaches to survival analysis, our approach is able not only to achieve superior ordering performance, but also to predict the survival time very well. Experiments show a significant performance improvement when the majority of the training data is censored. Experimental results on several survival analysis datasets demonstrate that SVCR is very competitive against classical survival analysis models.

It is well known that occurrence counts of words in documents are often modeled poorly by standard distributions like the binomial or Poisson. Observed counts vary more than simple models predict, prompting the use of overdispersed models like Gamma-Poisson or Beta-binomial mixtures as robust alternatives. Another deficiency of standard models is due to the fact that most words never occur in a given document, resulting in large amounts of zero counts. We propose using zero-inflated models for dealing with this, and evaluate competing models on a Naive Bayes text classification task. Simple zero-inflated models can account for practically relevant variation, and can be easier to work with than overdispersed models.

Voicemail is not like email. Even such basic information as the name of the caller/sender or a phone number for returning calls is not represented explicitly and must be obtained from message transcripts or other sources. We discuss techniques for doing this and the challenges these tasks present.

Semantic interpretation for limited-domain spoken dialogue systems often amounts to extracting information from utterances. For a system that provides movie showtime information, queries are classified along four dimensions: question type, and movie titles, towns and theaters that were mentioned. Simple heuristics suffice for constructing highly accurate classifiers for the latter three attributes; classifiers for the question type attribute are induced from data using features tailored to spoken language phenomena. Since separate classifiers are used for the four attributes, which are not independent, certain errors can be detected and corrected, thus increasing robustness. 1.