Main

Adaptation of Speaking Style for Expressive Speech Synthesis


Recommandations


  • Check if the Flash plugin works correctly and the sound level is properly set.
  • Use headphones or earphones.
  • Do the test in a quiet place.
  • Before running the test, do not hesitate to send me an e-mail if you have any question.

The best performance is currently owned by a native French speaker expert in speech processing, with a rate of 77% correct identification. Non French speakers perform surprisingly well, especially musicians who performed a substantial rate of 62% correct identification. --- Last update: 2011, january 11th, 20.30.

Framework


Thank you for participating ! This experiment won't take you more than 15 minutes .

This study aims at evaluating speaking style adaptation in speech synthesis.

Each speaker has his own speaking style which constitutes his voice signature, and a part of his identity. Nevertheless, a speaker continuously adapt his speaking style according to specific communication situations.

Or, each situational context determines a specific mode of production associated with it - a genre - which is defined by a set of conventions of form and content that is shared among all of its productions. In particular, a specific discourse genre relate to a specific speaking style. .

Instructions


You are going to listen to some speech samples that have been filtered in order to focus on their prosodic dimension.

For each sample, you have to associate it with one of the proposed speaking style as described bellow.

  • First listen to the samples by clicking on the audio player (do not listen to a sample more than three times).
  • Then select a speaking style as quickly as possible. As you may have some trouble in choosing one and only one speaking style, we offer you several possibilities:
    • if you are sure of your choice, select only one speaking style;
    • if you hesitate between two speaking styles, select both;
    • if the speaking style sounds neutral, or if none of the speaking styles corresponds to that synthesized, select the "?" label.
  • To reset a file and its related choices, just click again on the audio player.

Symbols


There are 4 four speaking styles for that experiment.

Here is a description and the symbols related to the different speaking styles.

  • P = political (TV new year's speech)
  • J = journalistic (radio press review and chronicle)
  • S = sport commentary (soccer)
  • M = mass (church service, christian sermon)
  • N = the speaking style is neutral

Pre-Experiment


In order to kindly remind you how sound the speaking styles to identify, and to test your identification ability, please start the experiment with the following real speech samples.
You can use this pre-experiment to familiarize with the experiment procedure.

File Audio P J S M N
1

2

3

4

5

6

7

8

Main Experiment


Here is a couple of synthesized utterances of the original speaker to be adapted, namely the French actor André Dussolier.

speaker

Now, the speaking style of the original speaker has been transformed torwards the different speaking styles. For each utterance, you have to indentify the speaking style in which the original speaker has been transformed. The text is neutral, so you must not consider its content to make your decision.

We strongly recommend you to process the experiment in several steps:

  • First, listen to all of the samples prior to start the experiment;.
  • Then, select the speaking style for the cases there is no doubt;
  • Precise iteratively your selections until all samples have been decided.

File Audio P J S M N
1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

File Audio P J S M N

Some information about you


Listening conditions headphones other

Langage french native french speaking other

Expertise (are you familiar to a domain related to speech, such as speech processing, linguistic, ... ?) yes no

Expertise Domain (if you are expert, what is your precise domain: speech, linguistic, music, ...)

Age

Mail address and name (strongly recommanded if you've participated to the previous experiment)

Comments

Sending


Before sending please check that all the files of the main experiment got an answer.

Click on "send your answers" to get your result !! You can redo the test as many time as you want to improve your score.

All recordings are Ircam's property.