Documents
Resources
Learning Center
Upload
Plans & pricing Sign in
Sign Out

Acoustical Analysis of Emotional Speech in Standard Basque for

VIEWS: 6 PAGES: 11

									    Subjective evaluation of an
    emotional speech database
            for Basque


Aholab Signal Processing Laboratory – University of the Basque Country

Authors: I. Sainz, I. Saratxaga, E. Navas, I. Hernáez, J. Sanchez, I. Luengo, I. Odriozola
                    Index

   Introduction

   Corpus Design and Recording

   Evaluation Process

   Results

   Conclusions & Work in Progress
                        Introduction

   Speech Synthesis
     Intelligibility
     Naturalness
     Variability

   Emotional Speech
     Prosody
     Acoustic Features
     Corpus Based Tecniques
           Implicit Acoustic modeling
     Corpus Design and Recording

   Aim
     Emotional Corpus-based TTS
     Prosodic and acoustic analysis

   Recording emotional speech
     Spontaneous
     Elicited
     Acted

   2 speakers, 702 sentences
       1 hour per emotion
     Corpus Design and Recording

   Big 6 Emotions
             Evaluation Process

   Subjetive evaluation campaign
     30 stimuli per actor
     Forced choice




   20 Subjects
       11 native speakers
                      Results
                 Confusion Matrix




   Recognition average 76.6%
       Far above chance level 17%
            Results
Confusion Matrices(Actress & Actor)



                               Average
                               75.83%




Average
76.50%
                            Results
                    Effects of Listeners

   Student’s t-test
       Determine if the hypothesis is true
   Groups (recognition rate)
       Women (72.78%) Vs Men (77.62%)
            Not significant (95% confidence interval)
       Native speakers (77.12%) Vs Non-native (75%)
            Not significant (95% confidence interval)
       1st Half of the test (72.64%) Vs 2nd Half (79.7%)
          t=2.85  p=0.0044 < 0.05 Significant!
          Recognition rate increase of 7%

              – Almost constant for all groups
Conclusions & Work in progress

   Valid resource of emotional speech for
    Basque language
       Emotions readily recognized for both actors
   Study and modeling of emotional speech
   Corpus based TTS with emotions
       Neutral already developed
    Subjective evaluation of an
    emotional speech database
            for Basque


Aholab Signal Processing Laboratory – University of the Basque Country

Authors: I. Sainz, I. Saratxaga, E. Navas, I. Hernáez, J. Sanchez, I. Luengo, I. Odriozola

								
To top