View
218
Download
0
Category
Tags:
Preview:
Citation preview
The Role of F0 in the Perceived Accentedness of L2
Speech
Mary Grantham O’Brien
Stephen Winters
GLAC-15, Banff, Alberta
May 1, 2009
Thank you
• Roswita Dressler
• Annika Orich
• Valerie Haberl
• Robert MacDonald
• Silke Weber
• Andreas Berkefeld
Motivation• What effect does non-native prosody have on the
perceived accentedness and intelligibility of speech?
• Production studies: perceptual interference between L1 and L2 (Holm 2008; McAllister 1997).
• Possible to rate accentedness in unfamiliar foreign language (Major 2007)
• Perceptual studies: intelligibility, comprehensibility and accentedness (e.g., Munro et al. 2006; Kennedy & Trofimovich 2008)
Intelligibility of L2 speech
• “the extent to which a speaker’s utterance is actually understood” (Munro et al. 2006, p. 112)
• measured through listener transcriptions• Accented speech more intelligible if listeners
share L1 of speaker (Kennedy & Trofimovich 2008)
• Positively affected by intonation manipulation.
Role of intonation in perception of foreign accent
• Differs according to study and L1-L2 pairings• Segments• Duration
Cross-linguistic intonation
• Languages differ both in average pitch height and range of frequencies (e.g., Braun 1994; Ladd 1996)
• German vs. English intonation• Greater pitch range in English than in German
(Eckert & Laver 1994; Gibbon 1998; Mennen 2007)
• Bilingual intonation• Bilinguals adjust intonation, depending on
language being spoken (Mennen 2007)
The current study• Transplantation of both native and non-native intonation patterns onto the segments of the other language.
• German and English speakers
• Q1: Does mapping non-native intonation onto native segments affect perceived accentedness and intelligibility?
• Q2: Does mapping native intonation onto non-native segments affect perceived accentedness and intelligibility?
• Also: listeners included L2 German learners at a variety of proficiency levels.
Methodology: Stimuli• 24 German and English sentences
• 12 declarative
• 6 yes/no questions
• 6 WH-questions
• Produced by:
• L1 English/L2 German (female) speaker
• L1 German/L2 English (female) speaker
Stimulus Conversion• German speaker prosody transplanted onto English speaker segments,
• and vice versa…
• for productions in both languages.
• PSOLA-based manipulation algorithm developed by Mareüil & Vieru-Dimulescu (2006)
• Pink noise added to all stimuli for intelligibility test.
• 0 dB SNR for English stimuli
• +5 dB SNR for German stimuli
Stimuli Samples
• Unedited English:
• Converted English:
• Unedited German:
• Converted German:
Clear In Noise
Methodology: Tasks1. Intelligibility
• Listeners heard natural + converted sentences in noise
• Typed in answers to question “What did the speaker say?”
2. Accent rating
• Listeners heard natural + converted sentences in clear
• Rated accentedness on a scale from 0 to 6
• (0 = “no accent”, 6 = “most accent”)
• Two counterbalanced blocks: German and English
Methodology: Listeners• Subjects: 22 native listeners of Canadian English
• 14 males, 8 females
• Aged 20-25 (mean: 22 years)
• Beginning to advanced learners of German
• Levels A2-C1 on Common European Framework of Reference for Languages (mostly levels B1 and B2)
Results: AccentednessPerceived Accentedness
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
ef - natural ef - gF0 gf - natural gf - eF0
Stimulus Type
Accent Rating (0-6)
English German
Figure 3
Results: IntelligibilityIntelligibility
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
ef - natural ef - gF0 gf - natural gf - eF0
Stimulus Type
% Words Correctly Identified
English German
Figure 4
Results: German Accent Ratings by Listener Group
German Accent Ratings, by Listener Group
0.00
0.50
1.00
1.50
2.00
2.50
3.00
3.50
ef - natural ef - gF0 gf - natural gf - eF0
Stimulus Type
Accent Rating (0-6)
High Proficiency Low ProficiencyFigure 5
Results: German Intelligibility by Listener Group
German Intelligibility, by Listener Group
0%
10%
20%
30%
40%
50%
60%
ef - natural ef - gF0 gf - natural gf - eF0
Stimulus Type
% Words Correctly Identified
High Proficiency Low ProficiencyFigure 6
Results: Summary• Accentedness:
1. F0 transplantation decreases perceived accentedness of natively produced sentences…
2. But does not improve perceived accentedness of non-natively produced sentences.
• Intelligibility:
1. English sentences more intelligible than German
2. English speaker more intelligible than German in both language conditions
3. F0 manipulation reduces intelligibility in all conditions
Results: Summary• Low proficiency German learners based accent ratings more on prosodic features;
• High proficiency learners’ ratings based more on natural vs. manipulated.
• More top-down processing in case of high proficiency learners?
Conclusions• Prosodic characteristics do play a role in the perception of accentedness (or lack thereof).
• true for both L1 and L2 perception.
• L2 listeners better understood L2 speaker in adverse listening conditions.
• Note: “Interlanguage speech intelligibility benefit” (Bent & Bradlow, 2003)
• Note: PSOLA-based manipulation induced unnaturalness into speech stimuli.
• possible limitations of synthesis algorithm
Future research• Possible control: compare manipulated sentences with manipulated sentences.
• Also: perception of naturally produced F0 contours in both conditions (by phonetically trained speaker)
• Other directions: test native listeners of German
• Test broader range of speakers
• Test speakers of other languages with clearer prosodic differences
• Pitch Accent languages (French, Japanese)
• Tone languages (Chinese)
Accent Rating Screen
Results: NaturalnessPerceived Naturalness
0
0.5
1
1.5
2
2.5
3
3.5
4
ef - natural ef - gF0 gf - natural gf - eF0
Stimulus Type
Naturalness Rating (0-6)
English German
Recommended