Speech Synthesis Research Research Articles

RoomWeaver and Digitract: Two digital waveguide mesh acoustic modeling reseach tools

RoomWeaver is a digital waveguide mesh (DWM) development environment for virtual acoustic spaces. Scripting is used to define room geometry, object, surface and material properties, together with source and receiver parameters. Different mesh types and topologies are accommodated through a user-extensible plug-in architecture with the selected DWM grown to fit the defined room at simulation run-time. Visualization options allow mesh behavior and wave propagation to be observed in 2D or 3D. Excitation signals can be selected accordingly and resultant room impulse responses can be obtained in mono, stereo, binaural or surround formats. Digitract is a speech synthesis research tool that uses 1D and 2D digital waveguide models to simulate the human vocal tract. 1D simulation is already well established and Digitract allows direct comparison with 2D DWM models that offer improved control over formant bandwidths and greater naturalness in synthesized speech. Mesh types and topology can be defined together with reflection values at lip, glottis, and, in the 2D case, inner wall boundaries. Vibrato and pitch-slide can be introduced to the input and dipthongs and tripthongs can also be synthesized. 3D graphical feedback illustrates the behavior of the acoustic pressure waves present within the tract model.

Measuring speech quality for text-to-speech systems: development and assessment of a modified mean opinion score (MOS) scale

The quality of text-to-speech systems can be effectively assessed only on the basis of reliable and valid listening tests to assess overall system performance. A mean opinion scale (MOS) has been the recommended measure of synthesized speech quality [ITU-T Recommendation P.85, 1994. Telephone transmission quality subjective opinion tests. A method for subjective performance assessment of the quality of speech voice output devices]. We assessed this MOS scale and developed and tested a modified measure of speech quality. This modified measure has new items specific to text-to-speech systems. Our research was motivated by the lack of clear evidence of the conceptual content of as well as the psychometric properties of the MOS scale. We present conceptual arguments and empirical evidence for the reliability and validity of a modified scale. Moreover, we employ state of the art psychometric techniques such as confirmatory factor analysis to provide strong tests of psychometric properties. This modified scale is better suited to appraise synthesis systems since it includes items that are specific to the artifacts found in synthesized speech. We believe that the speech synthesis research communities will find this modified scale a better fit for listening tests to assess synthesized speech.

Statistical prosodic modeling: from corpus design to parameter estimation

The increasing availability of carefully designed and collected speech corpora opens up new possibilities for the statistical estimation of formal multivariate prosodic models. At Apple Computer, statistical prosodic modeling exploits the Victoria corpus, created to broadly support ongoing speech synthesis research and development. This corpus is composed of five constituent parts, each designed to cover a specific aspect of speech synthesis: polyphones, prosodic contexts, reiterant speech, function word sequences, and continuous speech. This paper focuses on the use of the Victoria corpus in the statistical estimation of duration and pitch models for Apple's next-generation text-to-speech system in Macintosh OS X. Duration modeling relies primarily on the subcorpus of prosodic contexts, which is instrumental to uncover empirical evidence in favor of a piecewise linear transformation in the well-known sums-of-products approach. Pitch modeling relies primarily on the subcorpus of reiterant speech, which makes possible the optimization of superpositional pitch models with more accurate underlying smooth contours. Experimental results illustrate the improved prosodic representation resulting from these new duration and pitch models.

Approximate model for sound generation due to unsteady flows in pipes

An approximate model for the generation of sound due to unsteady vortical flow in pipes of varying cross-sectional area is described. This model, which is being used in speech synthesis research, is based upon Howe’s acoustic analogy. This formulation allows the sound generation to be specified in terms of properties of the unsteady behavior of a vortical flow and the potential flow solution for the pipe that would exist in the absence of vortical inhomogeneities. An approximate model for vorticity formation and evolution is presented, and the potential flow solution is obtained from the axial duct shape. Computed estimates of the time evolution of the sound generated by a confined jet passing through a pipe constriction are presented. These results are compared to experimental data. [Research supported by NSF/ARPA IRI-9314946 and ARPA DAST 63-93-C-0064.]

Speech technology in 2001: new research directions.

Research in speech recognition and synthesis over the past several decades has brought speech technology to a point where it is being used in "real-world" applications. However, despite the progress, the perception remains that the current technology is not flexible enough to allow easy voice communication with machines. The focus of speech research is now on producing systems that are accurate and robust but that do not impose unnecessary constraints on the user. This chapter takes a critical look at the shortcomings of the current speech recognition and synthesis algorithms, discusses the technical challenges facing research, and examines the new directions that research in speech recognition and synthesis must take in order to form the basis of new solutions suitable for supporting a wide range of applications.

A brief history of synthetic speech

This paper retraces, in an informal way, some of the history of speech synthesis and speech research from Von Kempelen's speaking machine to linear prediction.

Synthesis: Modeling variability and constraints

This paper discusses some important topics in current speech synthesis research. Modeling of speaker characteristics and emotions are used as an example of new trends in the speech synthesis field. The relation to speech recognition research is emphasized. New methods such as automatic learning and the use of new analysis techniques are also discussed.

Interactive utterance modeling with Delta for research in speech synthesis and speech perception

Productive research in speech synthesis and perception requires the ability to readily test the effect of selected modifications to an utterance (e.g., shortening of voice onset time). With the Delta system software [J. Acoust. Soc. Am. Suppl. 1 85, S124 (1989)], “multistream” utterance representations (deltas) consisting of coordinated user‐defined acoustic or articulatory units can be built and manipulated interactively, and the effects of modifications tested through synthesis. To facilitate experimentation in the acoustic domain, a set of Delta procedures have been developed for interactively generating an initial delta for a Klatt synthesizer in any natural language. This delta contains a phoneme stream, streams for synthesizer parameters, and “slots” for perceptually important acoustic events such as stop bursts and formant transitions. With such a delta, the experimenter can easily manipulate acoustic values and their timing. Such interactive utterance modeling is being used to develop synthesis rules for several English dialects and for Hindi, and soon the approach will be extended to other languages. The system will be demonstrated. [Work supported by U.S. Dept. of Education and NIH.]

Formant synthesis: Technique to account for source/tract interaction

Research in speech synthesis is presently concerned with methods for improving intelligibility and naturalness. But most work to date has dealt with intelligibility, i.e., the synthesized speech is intelligible but monotone and machinelike in quality. Although this type of speech is useful in some applications, the users of speech synthesizers are demanding a more natural sounding synthesized speech. We are addressing this issue in our research. In particular we are studying the influence of the glottal source function on the production of synthesized speech. Our data base consists of inverse filtered speech, ultra‐high‐speed laryngeal films, and the electroglottograph waveforms, all temporally synchronized in our experiments. We use the experimentally derived source waveforms obtained from these various method as the excitation for a serial/parallel Klatt formant synthesizer to synthesize sentences. We then evaluate the naturalness of this synthetic speech by listening tests. Our goal is to rank order th...

Synthesis of an Optical Sound Spectrograph

The basic theory of a large space bandwidth signal analysis by means of a two-dimensional optical processor is discussed and the synthesis of a near real-time optical sound spectrograph is presented. In the current state-of-the-art, the realization of an optical sound spectrograph is feasible. It is also concluded that the optical sound spectrograph may offer a broader frequency range of analysis and flexibility in spectrum data handling. The former may be important in ultrasonic signal analysis; the latter may be important in future research in automatic speech recognition, synthesis, and translation.

Experimental Conversational Computer System

This paper describes an experimental system with speech input and output providing a problem domain for speech analysis and synthesis research. Speech input to the computer is provided using a system developed at Stanford (Reddy and Vicens, “Spoken Message Recognition by a Computer,” to be published). Time-domain compressed speech [D. R. Reddy, J. Acoust. Soc. Amer. 44, 391(A) (1968)] is used for voice response from the computer. If a query is not one of the 25 or so messages the system can recognize, it responds with “Please repeat,” or some such. If the input is recognized, then the response is a random but relevant message chosen from about 100 messages. This system was developed not so much to talk to machines but rather to study some aspects of man-machine voice communication such as the effect of (a) a delayed reply on a human speaker, (b) intonationless speech that can result from concatenating isolated words to form a sentence, (c) the increase of vocabulary on computer speed and storage requirements, and (d) improvements to the analysis or synthesis algorithms on the over-all performance of the system.

Thirty Years of Vocoder Research

A program of active research on the vocoder was initiated in July 1934. The first public demonstration of the vocoder was at the Harvard Tercentenary in September 1936. Since then, there have been studies of the underlying nature of speech and the recognition of speech-sound elements in transforming from the spoken to the written word or setting up other automatic voice operation. There has also been much research directed toward straightforward bandwidth compression and such compression as an aid to the attainment of other goals. The entire field of research in speech analysis and/or synthesis is reviewed, including a discussion of limitations and requirements. A demonstration tape is played from an improved 2400-bit/sec digitized vocoder.