Baseline System for Automatic Speech Recognition with French GlobalPhone Database

Milos Cernak, Sandrine Revaz
Idiap, 2012
Report or working paper

Abstract

This report presents one month trainee work on development of French Automatic Speech Recognition ASR system using a french part of multilingual database GlobalPhone_FR. The purpose of this report is to explain and give results of the training and testing of the ASR with this specific database. Two different methods are presented, the Hidden Markov Model (HMM) with MFCC/PLP features and tandem features from Multilayer Perceptron (MLP) phone posteriors. The report presents data preparation for GlobalPhone_FR ASR training, and compares the two different approaches. Word recognition accuracy achieved with MFCC features is 71.46% and the tandem features with 3-layer MLP improved the accuracy to 72.15%. We interpret this result as a baseline for the GlobalPhone_FR database.

Official source

https://infoscience.epfl.ch/record/192600?ln=en

About this result

This page is automatically generated and may contain information that is not correct, complete, up-to-date, or relevant to your search query. The same applies to every other page on this website. Please make sure to verify the information with EPFL's official sources.

Related concepts

Related publications

Related publications (19)

SVR vs MLP for Phone Duration Modelling in HMM-based Speech Synthesis

Philip Neil Garner, Pierre-Edouard Jean Charles Honnet

In this paper we investigate external phone duration models (PDMs) for improving the quality of synthetic speech in hidden Markov model (HMM)-based speech synthesis. Support Vector Regression (SVR) and Multilayer Perceptron (MLP) were used for this task. SVR and MLP PDMs were compared with the explicit duration modelling of hidden semi-Markov models (HSMMs). Experiments done on an American English database showed the SVR outperforming the MLP and HSMM duration modelling on objective and subjective evaluation. In the objective test, SVR managed to outperform MLP and HSMM models achieving 15.3% and 25.09% relative improvement in terms of root mean square error (RMSE) respectively. Moreover, in the subjective evaluation test, on synthesized speech, the SVR model was preferred over the MLP and HSMMmodels, achieving a preference score of 35.93% and 56.30%, respectively.

2014

In this paper we investigate external phone duration models (PDMs) for improving the quality of synthetic speech in hidden Markov model (HMM)-based speech synthesis. Support Vector Regression (SVR) and Multilayer Perceptron (MLP) were used for this task. SVR and MLP PDMs were compared with the explicit duration modelling of hidden semi-Markov models (HSMMs). Experiments done on an American English database showed the SVR outperforming the MLP and HSMM duration modelling on objective and subjective evaluation. In the objective test, SVR managed to outperform MLP and HSMM models achieving 15.3% and 25.09% relative improvement in terms of root mean square error (RMSE) respectively. Moreover, in the subjective evaluation test, on synthesized speech, the SVR model was preferred over the MLP and HSMM models, achieving a preference score of 35.93% and 56.30%, respectively.

Idiap2014

In this paper, we present a new approach towards high performance speech/music discrimination on realistic tasks related to the automatic transcription of broadcast news. In the approach presented here, the (local) Probability Density Function (PDF) estimators trained on clean, microphone, speech (as used in a standard large vocabulary speech recognition system) are used as a channel model at the output of which the entropy and

dynamism'' will be measured and integrated over time through a 2-state (speech and and non-speech) hidden Markov model (HMM) with minimum duration constraints. Indeed, in the case of entropy, it is clear that, on average, the entropy at the output of the local PDF estimators will be larger for speech signals than non-speech signals presented at their input. In our case, local probabilities will be estimated from an multilayer perceptron (MLP) as used in hybrid HMM/MLP systems, thus guaranteeing the use of

real'' probabilities in the estimation of the entropy. The 2-state speech/non-speech HMM will thus take these two dimensional features (entropy and ``dynamism'') whose distributions will be modeled through (two-dimensional) multi-Gaussian densities or an MLP, whose parameters are trained through a Viterbi algorithm.\ Different experiments, including different speech and music styles, as well as different (a priori) distributions of the speech and music signals (real data distribution, mostly speech, or mostly music), will illustrate the robustness of the approach, always resulting in a correct segmentation performance higher than 90%. Finally, we will show how a confidence measure can be used to further improve the segmentation results, and also discuss how this may be used to extend the technique to the case of speech/music mixtures.

IDIAP2001