Skip to Main content Skip to Navigation
Conference papers

Independent and Automatic Evaluation of Speaker-Independent Acoustic-to-Articulatory Reconstruction

Maud Parrot 1 Juliette Millet 2, 1 Ewan Dunbar 2, 1
1 CoML - Apprentissage machine et développement cognitif
LSCP - Laboratoire de sciences cognitives et psycholinguistique, Inria de Paris
Abstract : Reconstruction of articulatory trajectories from the acoustic speech signal has been proposed for improving speech recognition and text-to-speech synthesis. However, to be useful in these settings, articulatory reconstruction must be speakerindependent. Furthermore, as most research focuses on single, small data sets with few speakers, robust articulatory reconstruction could profit from combining data sets. Standard evaluation measures such as root mean squared error and Pearson correlation are inappropriate for evaluating the speakerindependence of models or the usefulness of combining data sets. We present a new evaluation for articulatory reconstruction which is independent of the articulatory data set used for training: the phone discrimination ABX task. We use the ABX measure to evaluate a bi-LSTM based model trained on three data sets (14 speakers), and show that it gives information complementary to standard measures, enabling us to evaluate the effects of data set merging, as well as the speaker independence of the model.
Document type :
Conference papers
Complete list of metadata

https://hal.archives-ouvertes.fr/hal-03087264
Contributor : Ewan Dunbar <>
Submitted on : Wednesday, December 23, 2020 - 4:04:20 PM
Last modification on : Thursday, July 1, 2021 - 5:54:07 PM

File

Interspeech_2020__articulatory...
Files produced by the author(s)

Identifiers

  • HAL Id : hal-03087264, version 1

Citation

Maud Parrot, Juliette Millet, Ewan Dunbar. Independent and Automatic Evaluation of Speaker-Independent Acoustic-to-Articulatory Reconstruction. Interspeech 2020 - 21st Annual Conference of the International Speech Communication Association, Oct 2020, Shanghai / Virtual, China. ⟨hal-03087264⟩

Share

Metrics

Record views

88

Files downloads

120