The IBM trainable speech synthesis system

Donovan, Robert E.; Eide, Ellen

doi:10.21437/icslp.1998-10

Cited by 46 publications

(9 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…It is generally known that up to three formants are required to synthesize intelligible speech or reconstruct lexical information [6]. Our approach to privacy, motivated by source-filter model of speech production, is based on adaptively filtering the spectral peaks.…”

Section: Analysis Of Residual Featuresmentioning

confidence: 99%

“…A prediction order of 8 seems appropriate since the first two formants are important for synthesizing intelligible speech [6]. Although the 8 th order residual signal can be intelligible, informal experiments synthesizing speech from the MFCC representation of an 8 th order residual does not sound intelligible.…”

Section: Analysis Of Lp Ordermentioning

confidence: 99%

“…An increase from 2 to 6 results in a drop of 1.6% in the MDM case. This could be due to the loss of the first formant, which carries more linguistic information [6]. For LP orders between 8 to 12, an increase in the LP order results in a bigger drop in performance.…”

Section: Effect Of Prediction Order On Diarizationmentioning

confidence: 99%

“…Drawing motivation from the source-filter model, our approach to privacy is based on adaptively filtering formants required to synthesize intelligible speech [6]. To this end, we present linear prediction (LP) residual as a privacy-sensitive feature.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

LP residual features for robust, privacy-sensitive speaker diarization

Parthasarathi¹,

Bourlard²,

Gática-Pérez³

2011

Interspeech 2011

View full text Add to dashboard Cite

We present a comprehensive study of linear prediction residual for speaker diarization on single and multiple distant microphone conditions in privacy-sensitive settings, a requirement to analyze a wide range of spontaneous conversations. Two representations of the residual are compared, namely real-cepstrum and MFCC, with the latter performing better. Experiments on RT06eval show that residual with subband information from 2.5 kHz to 3.5 kHz and spectral slope yields a performance close to traditional MFCC features. As a way to objectively evaluate privacy in terms of linguistic information, we perform phoneme recognition. Residual features yield low phoneme accuracies compared to traditional MFCC features.

show abstract

Section: Analysis Of Residual Featuresmentioning

confidence: 99%

Section: Analysis Of Lp Ordermentioning

confidence: 99%

Section: Effect Of Prediction Order On Diarizationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

LP residual features for robust, privacy-sensitive speaker diarization

Parthasarathi¹,

Bourlard²,

Gática-Pérez³

2011

Interspeech 2011

View full text Add to dashboard Cite

show abstract

“…The principle of speech synthesis by HMM was first introduced by Tokuda et al [18] for acoustic speech synthesis and extended to audiovisual speech by the HTS working group [19]. Note that the idea of exploiting HMM capabilities for grasping essential sound characteristics for synthesis was also promoted by various authors such as Giustiniani and Pierucci [20] and Donovan [21]. The HMM-trajectory synthesis technique comprises training and synthesis parts (see [22,23] for details).…”

Section: The Trajectory Formation Systemmentioning

confidence: 99%

Lip-Synching Using Speaker-Specific Articulation, Shape and Appearance Models

Bailly¹,

Govokhina²,

Elisei³

et al. 2009

EURASIP Journal on Audio, Speech, and Music Processing

View full text Add to dashboard Cite

We describe here the control, shape and appearance models that are built using an original photogrammetric method to capture characteristics of speaker-specific facial articulation, anatomy, and texture. Two original contributions are put forward here: the trainable trajectory formation model that predicts articulatory trajectories of a talking face from phonetic input and the texture model that computes a texture for each 3D facial shape according to articulation. Using motion capture data from different speakers and module-specific evaluation procedures, we show here that this cloning system restores detailed idiosyncrasies and the global coherence of visible articulation. Results of a subjective evaluation of the global system with competing trajectory formation models are further presented and commented.

show abstract