Neural network based generation of fundamental frequency contours

Scordilis, Michael S.; Gowdy, J.N.

doi:10.1109/icassp.1989.266404

Cited by 31 publications

(17 citation statements)

References 3 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our model has the advantage over these purely rule-based models in that it can be automatically trained. Since the structure is based in linguistic theory, it also covers a broader range of prosodic characteristics than do other automatically-trainable models, e.g., [38]. A potential advantage of the dynamical system model is that it can handle vector processes, allowing for joint modeling of energy and .…”

Section: Discussionmentioning

confidence: 99%

“…For example, consider the following relatively simple case with parameter tying only in the observation equation: (38) In this case, both and have the data broken up into regions according to the factor , but for the data must be further partitioned according to the factor . If can take on two values ( and ) then the data is broken up to define two parameters and for every .…”

Section: Parameter Tyingmentioning

confidence: 99%

See 1 more Smart Citation

A dynamical system model for generating fundamental frequency for speech synthesis

Ross

Ostendorf

1999

IEEE Trans. Speech Audio Process.

View full text Add to dashboard Cite

Higher quality speech synthesis is required for widespread use of text-to-speech (TTS) technology, and prosody is one component of synthesis technology with the greatest need for improvement. This paper describes a new approach to generation of two important cues to prosodic patterns-fundamental frequency (F 0 ) and energy contours-given symbolic prosodic labels and text. Specifically, the approach represents vectors of F 0 and energy with a dynamical system model, which allows automatic estimation of parameters from labeled speech. Parameters at different time scales in the model are structured to capture segment, syllable, phrase and discourse level effects based on linguistic research. F0 generation experiments with the dynamical system model show improved synthetic speech quality over the hybrid target/filter approach.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Parameter Tyingmentioning

confidence: 99%

A dynamical system model for generating fundamental frequency for speech synthesis

Ross

Ostendorf

1999

IEEE Trans. Speech Audio Process.

View full text Add to dashboard Cite

show abstract

“…Among several database-driven methods, CART and neural network models are more popular (Breiman et al 1984;Dusterhoff et al 1999;Goubanova & King 2008;Cosi et al 2001;Tesser et al 2004;Vainio & Altosaar 1998;Vegnaduzzo 2003). Several models based on neural network principles are described in the literature for predicting the intonation patterns of syllables in continuous speech (Scordilis & Gowdy 1989;Vainio & Altosaar 1998;Vainio 2001;Sonntag et al 1997;Buhmann et al 2000;Hwang & Chen 1994). Scordilis & Gowdy (1989) used neural networks in a parallel and distributed manner to predict the average F 0 value for each phoneme, and the temporal variations of F 0 within a phoneme.…”

Section: Input Layermentioning

confidence: 99%

“…Several models based on neural network principles are described in the literature for predicting the intonation patterns of syllables in continuous speech (Scordilis & Gowdy 1989;Vainio & Altosaar 1998;Vainio 2001;Sonntag et al 1997;Buhmann et al 2000;Hwang & Chen 1994). Scordilis & Gowdy (1989) used neural networks in a parallel and distributed manner to predict the average F 0 value for each phoneme, and the temporal variations of F 0 within a phoneme. The network consists of two levels: macroscopic and microscopic levels.…”

Section: Input Layermentioning

confidence: 99%

Role of neural network models for developing speech systems

Rao

2011

Sadhana

View full text Add to dashboard Cite

This paper discusses the application of neural networks for developing different speech systems. Prosodic parameters of speech at syllable level depend on positional, contextual and phonological features of the syllables. In this paper, neural networks are explored to model the prosodic parameters of the syllables from their positional, contextual and phonological features. The prosodic parameters considered in this work are duration and sequence of pitch (F 0 ) values of the syllables. These prosody models are further examined for applications such as text to speech synthesis, speech recognition, speaker recognition and language identification. Neural network models in voice conversion system are explored for capturing the mapping functions between source and target speakers at source, system and prosodic levels. We have also used neural network models for characterizing the emotions present in speech. For identification of dialects in Hindi, neural network models are used to capture the dialect specific information from spectral and prosodic features of speech.

show abstract

“…Model-based (e.g., regression trees, HMMs, neural networks) or sample-based (e.g., vector quantization, contour selection) mapping tools are then used to achieve the best phonetic prediction according to a distance metric, generally Root Mean Square (RMS) error. Prediction is generally performed with separate trainable models for f0 (Ljolje and Fallside 1986;Scordilis and Gowdy 1989;Sagisaka 1990;Traber 1992), for phoneme durations (Klatt 1979;O'Shaughnessy 1981;Bartkova and Sorin 1987;Riley 1992;van Santen 1992) and, more recently, for intensity profiles (Trouvain, Barry et al 1998). With the development of corpus-based synthesis techniques and powerful mapping tools (Campbell 1992;van Santen 2002), multiparametric prosodic models (Mixdorff and Jokisch 2001;Tesser, Cosi et al 2004) now tend to use general-purpose and theory-neutral tools.…”

Section: Introductionmentioning

confidence: 99%

SFC: A trainable prosodic model

Bailly

Holm

2005

Speech Communication

View full text Add to dashboard Cite

This paper introduces a new model-constrained and data-driven system to generate prosody from metalinguistic information. This system considers the prosodic continuum as the superposition of multiple elementary overlapping multiparametric contours. These contours encode specific metalinguistic functions associated with various discourse units. We describe the phonological model underlying the system and the specific implementation made of that model by the trainable prosodic model described here. The way prosody is analyzed, decomposed and modelled is illustrated by experimental work. In particular, we describe the original training procedure that enables the system to identify the elementary contours and to separate out their contributions to the prosodic contours of the training data.

show abstract

Neural network based generation of fundamental frequency contours

Cited by 31 publications

References 3 publications

A dynamical system model for generating fundamental frequency for speech synthesis

A dynamical system model for generating fundamental frequency for speech synthesis

Role of neural network models for developing speech systems

SFC: A trainable prosodic model

Contact Info

Product

Resources

About