Murray F. Spiegel

Comprehensive assessment of the telephone intelligibility of synthesized and natural speech

Abstract This paper describes a monosyllabic corpus for use in testing the consonant intelligibility of synthesized speech. It differs from those used in other tests in that it spans a wide variety of English sounds and is thus useful for diagnosis as well as for comparative assessment. Some “standard” tests of intelligibility use restricted phonetic material, which is possibly easier to understand than a representative sample of English; thus, the results from those tests may not reflect the intelligibility of a wider sample of speech. For illustration, we present the results of a telephone comparison between a demisyllable synthesizer currently being developed at Bellcore (“ Orator ”), a commercially available phoneme-based synthesizer, and natural speech obtained from 2 talkers. The natural speech data can be used by other laboratories wishing to compare the consonant intelligibility of other synthesis systems to natural speech.

Proper name pronunciations for speech technology applications

This paper describes a 15-year research effort to improve the automatic pronunciation of proper names and details the issues involved in applying those pronunciations to speech synthesis and speech recognition.

Modeling duration adjustment with dynamic time warping

Dynamic time warping (DTW) is used to time-align frames in a naturally spoken monosyllabic word-a long-duration syllabic-with the corresponding frames in the same syllable excised from a polysyllabic word-a shorter-duration rendition of the syllable. The DTW path, referenced to the monosyllable, indicates which portions of the monosyllable are shortened (and by how much) to match the shorter version of the syllable in a polysyllabic word. An experiment is presented which indicates that most shortening occurs in the latter portion of the syllable and that DTW path slopes are correlated with a measure of the amount of formant movement in the syllable. For speech synthesis by rule, a formant-movement time function computed for each template in an inventory in combination with additional information like phonemic vowel identity and consonant position within a syllable can define a time-compression function for the template/.<<ETX>>

3rd IEEE workshop on interactive voice technology for telecommunications applications

This special issue of Speech Communication contains 12 invited lectures and papers out of the 34 presented at the 3rd IEEE Workshop on Interactive Voice Technology for Telecommunications ApplicaŽ . tions IVTTA . The IVTTA workshop was held at the AT&T Learning Center in Basking Ridge, NJ, on 30 September and 1 October 1996. The papers published here were selected by members of the Scientific Program Committee of the workshop; all have been revised and extended by their authors following peer review by major researchers from around the world. The workshop was the third in a series of biennial meetings sponsored by the Communications Society of the IEEE. One hundred sixty participants from 20 countries in Europe, the Americas, Asia and the Middle East attended this conference. Each workshop in the series has brought together application researchers who planned to conduct or recently conducted field trials of new applications of speech technologies. For the most recent workshop, areas of interest covered a wide range of topics Žmobile telephony, ASR algorithms, language modeling and dialog design, text-to-speech, speech en. hancement, and speech application platforms , and Ž applications banking, operator services, network control, call routing, travel informationrreservations, . message retrieval, and speech translation . This is the second special issue of Speech Communication dedicated to the topic of Interactive Voice Technology for Telecommunications Applications. Ž The first Speech Communication, Volume 17, . Numbers 3–4, November 1995 contained a selection of papers from IVTTA ’94. Comparing that special issue to this one, we note that progress in the field has been gradual but steady. Although many technical challenges remain, this special issue includes more examples of successful spoken language understanding systems, indicating that we can now handle more complex, natural speech-enabled interactions. The papers selected for this special issue are a representative sampling of the many topics covered in the IVTTA workshop. These papers reflect the processes involved in moving speech technologies onto platforms to build systems that provide useful and usable services. The first three papers, by Johnston et al., Kitai et al., and Perdue, present overviews of a variety of voice-enabled telecommunications applications in Europe, Japan and the United States. The next two papers focus on specific applications deployed in the U.S. public switched network, and remind us that fielded services are not static, but require enhancement and modification as the technology or the requirements of the service change over time. Cerf et al. report on the NYNEX voicedialing service, first deployed in 1993, and describe the ramifications of introducing improvements to the

Advances in human-interface engineering for reverse directory assistance (ACNA) services

After years of productive research, speech synthesis is now profitably automating services by answering queries via constrained dialogs, directly accessing individual computer databases, and speaking text created from disparate sources of information. Directory-based services, such as Automated Customer Name and Address (ACNA), requires synthesis with high intelligibility and name pronunciation accuracy. Current synthesis technology achieves those goals. However, even the best of current speech technology is not good enough to mindlessly “drop” into complex services. Customized directory preprocessing is still necessary to transform listing data, which commonly contains unconventional abbreviations, unlabeled acronyms, and scrambled word ordering, into a sentence suitable for synthesis. This article describes state-of-the-art directory preprocessing programs that have led to successful implementations for synthesis services in 2 major U.S. telephone companies (Ameritech and Bell Atlantic). Of course, the basic capabilities of the synthesizer, such as pronuncïation accuracy, speech quality and naturalness, play a large role. Efforts ensured locality terms were pronounced in accordance with local custom. Finally, for prompts and other fixed messages, this article describes experiments that determined whether the naturalness of recorded speech offsets the undesirable discontinuity between recorded and synthesized utterances.

Challenges in evaluating the intelligibility of text‐to‐speech.

Text‐to‐speech (TTS) technology imposes different constraints on intelligibility than those sufficient for the evaluation of other speech communication systems. For example, the newly revised standard S.2‐2009 explicitly excludes TTS from the speech communication systems it covers. Since there is no current standard appropriate for evaluating TTS intelligibility, the ASA Standards Bioacoustics (S3) working group on Text‐to‐Speech Technology (WG91) was formed with the initial goal of developing such standard. We describe several ways in which standard methods of testing speech intelligibility are unsuitable for TTS technology and outline our approach to overcoming these limitations. We present an overview of our proposed standard, which is currently nearing its final draft stages.