Phonetic segmentation of speech using STEP and t-SNE

Adriana Stan; Cassia Valentini-Botinhao; Mircea Giurgiu; Simon King

doi:10.1109/SPED.2015.7343105

Phonetic segmentation of speech using STEP and t-SNE

Adriana Stan, Cassia Valentini-Botinhao, Mircea Giurgiu, Simon King

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

This paper introduces a first attempt to perform phoneme-level segmentation of speech based on a perceptual representation - the Spectro Temporal Excitation Pattern (STEP) - and a dimensionality reduction technique - the t-Distributed Stochastic Neighbour Embedding (t-SNE). The method searches for the true phonetic boundaries in the vicinity of those produced by an HMM-based segmentation. It looks for perceptually-salient spectral changes which occur at these phonetic transitions, and exploits t-SNE's ability to capture both local and global structure of the data. The method is intended to be used in any language and it is therefore not tailored to any particular dataset or language. Results show that this simple approach improves segmentation accuracy of unvoiced phonemes by 4% within a 5 ms margin, and 5% at a 10 ms margin. For the voiced phonemes, however, accuracy drops slightly.

Original language	English
Title of host publication	2015 International Conference on Speech Technology and Human-Computer Dialogue (SpeD)
Editors	Corneliu Burileanu, Corneliu Rusu, Horia-Nicolai Teodorescu, Horia-Nicolai Teodorescu
Publisher	Institute of Electrical and Electronics Engineers
Number of pages	6
ISBN (Electronic)	9781467375603
DOIs	https://doi.org/10.1109/SPED.2015.7343105
Publication status	Published - 3 Dec 2015
Event	8th International Conference on Speech Technology and Human-Computer Dialogue - Bucharest, Romania Duration: 14 Oct 2015 → 17 Oct 2015 https://sped.pub.ro/archive/sped2015/index-1.html

Conference

Conference	8th International Conference on Speech Technology and Human-Computer Dialogue
Abbreviated title	SpeD 2015
Country/Territory	Romania
City	Bucharest
Period	14/10/15 → 17/10/15
Internet address	https://sped.pub.ro/archive/sped2015/index-1.html

Keywords / Materials (for Non-textual outputs)

HMM acoustic model
k-Means
phonetic segmentation
STEP
t-SNE

Access to Document

10.1109/SPED.2015.7343105Licence: All Rights Reserved

https://ieeexplore.ieee.org/document/7343105Licence: All Rights Reserved

Simple4All: Speech synthesis that improves through adaptive learning
King, S. & Renals, S.
EU government bodies
1/11/11 → 31/10/14
Project: Research

File

Cite this

@inproceedings{7febe60267ff47e6afe463aa79390721,

title = "Phonetic segmentation of speech using STEP and t-SNE",

abstract = "This paper introduces a first attempt to perform phoneme-level segmentation of speech based on a perceptual representation - the Spectro Temporal Excitation Pattern (STEP) - and a dimensionality reduction technique - the t-Distributed Stochastic Neighbour Embedding (t-SNE). The method searches for the true phonetic boundaries in the vicinity of those produced by an HMM-based segmentation. It looks for perceptually-salient spectral changes which occur at these phonetic transitions, and exploits t-SNE's ability to capture both local and global structure of the data. The method is intended to be used in any language and it is therefore not tailored to any particular dataset or language. Results show that this simple approach improves segmentation accuracy of unvoiced phonemes by 4% within a 5 ms margin, and 5% at a 10 ms margin. For the voiced phonemes, however, accuracy drops slightly.",

keywords = "HMM acoustic model, k-Means, phonetic segmentation, STEP, t-SNE",

author = "Adriana Stan and Cassia Valentini-Botinhao and Mircea Giurgiu and Simon King",

year = "2015",

month = dec,

day = "3",

doi = "10.1109/SPED.2015.7343105",

language = "English",

editor = "Corneliu Burileanu and Corneliu Rusu and Horia-Nicolai Teodorescu and Horia-Nicolai Teodorescu",

booktitle = "2015 International Conference on Speech Technology and Human-Computer Dialogue (SpeD)",

publisher = "Institute of Electrical and Electronics Engineers",

address = "United States",

note = "8th International Conference on Speech Technology and Human-Computer Dialogue, SpeD 2015 ; Conference date: 14-10-2015 Through 17-10-2015",

url = "https://sped.pub.ro/archive/sped2015/index-1.html",

}

Stan, A, Valentini-Botinhao, C, Giurgiu, M & King, S 2015, Phonetic segmentation of speech using STEP and t-SNE. in C Burileanu, C Rusu, H-N Teodorescu & H-N Teodorescu (eds), 2015 International Conference on Speech Technology and Human-Computer Dialogue (SpeD). Institute of Electrical and Electronics Engineers, 8th International Conference on Speech Technology and Human-Computer Dialogue, Bucharest, Romania, 14/10/15. https://doi.org/10.1109/SPED.2015.7343105

Phonetic segmentation of speech using STEP and t-SNE. / Stan, Adriana; Valentini-Botinhao, Cassia; Giurgiu, Mircea et al.
2015 International Conference on Speech Technology and Human-Computer Dialogue (SpeD). ed. / Corneliu Burileanu; Corneliu Rusu; Horia-Nicolai Teodorescu; Horia-Nicolai Teodorescu. Institute of Electrical and Electronics Engineers, 2015.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Phonetic segmentation of speech using STEP and t-SNE

AU - Stan, Adriana

AU - Valentini-Botinhao, Cassia

AU - Giurgiu, Mircea

AU - King, Simon

PY - 2015/12/3

Y1 - 2015/12/3

N2 - This paper introduces a first attempt to perform phoneme-level segmentation of speech based on a perceptual representation - the Spectro Temporal Excitation Pattern (STEP) - and a dimensionality reduction technique - the t-Distributed Stochastic Neighbour Embedding (t-SNE). The method searches for the true phonetic boundaries in the vicinity of those produced by an HMM-based segmentation. It looks for perceptually-salient spectral changes which occur at these phonetic transitions, and exploits t-SNE's ability to capture both local and global structure of the data. The method is intended to be used in any language and it is therefore not tailored to any particular dataset or language. Results show that this simple approach improves segmentation accuracy of unvoiced phonemes by 4% within a 5 ms margin, and 5% at a 10 ms margin. For the voiced phonemes, however, accuracy drops slightly.

AB - This paper introduces a first attempt to perform phoneme-level segmentation of speech based on a perceptual representation - the Spectro Temporal Excitation Pattern (STEP) - and a dimensionality reduction technique - the t-Distributed Stochastic Neighbour Embedding (t-SNE). The method searches for the true phonetic boundaries in the vicinity of those produced by an HMM-based segmentation. It looks for perceptually-salient spectral changes which occur at these phonetic transitions, and exploits t-SNE's ability to capture both local and global structure of the data. The method is intended to be used in any language and it is therefore not tailored to any particular dataset or language. Results show that this simple approach improves segmentation accuracy of unvoiced phonemes by 4% within a 5 ms margin, and 5% at a 10 ms margin. For the voiced phonemes, however, accuracy drops slightly.

KW - HMM acoustic model

KW - k-Means

KW - phonetic segmentation

KW - STEP

KW - t-SNE

U2 - 10.1109/SPED.2015.7343105

DO - 10.1109/SPED.2015.7343105

M3 - Conference contribution

AN - SCOPUS:84960895368

BT - 2015 International Conference on Speech Technology and Human-Computer Dialogue (SpeD)

A2 - Burileanu, Corneliu

A2 - Rusu, Corneliu

A2 - Teodorescu, Horia-Nicolai

PB - Institute of Electrical and Electronics Engineers

T2 - 8th International Conference on Speech Technology and Human-Computer Dialogue

Y2 - 14 October 2015 through 17 October 2015

ER -

Phonetic segmentation of speech using STEP and t-SNE

Abstract

Conference

Keywords / Materials (for Non-textual outputs)

Access to Document

Fingerprint

Projects

Simple4All: Speech synthesis that improves through adaptive learning

Cite this