See also a comparsion of compression algorithms

How efficient is speech?, R.J.J.H. van Son and Louis C.W. Pols, In preparation

Each button adds a new factor.

Vowel information: IL and IS

########

Vowel Duration and Contrast

Vowel Contrast and Duration

Corrected Means

Stress and Word alignment versus Prominence

Information structure and efficiency in speech production, R.J.J.H. van Son and Louis C.W. Pols, EUROSPEECH 2003, Geneva, Switserland

Each button adds a new factor.

Segmental information: IS

Duration

Center of Gravity

Correlation between Duration and IS

Correlation between Center of Gravity and IS

An Acoustic Model of Communicative Efficiency in Consonants and Vowels taking into Account Context Distinctiveness, R.J.J.H. van Son and Louis C.W. Pols, ICPhS 2003 at Barcelona

Analyses, using Normalized Correlations for ALL phonemes (excluding schwa) and Phoneme info corrected for Plain Text Context Distinctiveness and excluding all phonemes having a corrected Phoneme info value <= 1.5 bits

Phoneme information as a function of Manner of Articulation and position in the word (excludes Schwa)
Figure 1: Relation between Segmental Lexical information corrected for Plain Text Context Distinctiveness and the position in the word grouped by manner of articulation for comparison.

Vowels: Style versus Prominence
Figure 2 top: Normalized Correlation Plain Text Context Corrected Phoneme info and vowel duration: Style X Autoprominence (X Speaker; Phoneme; Stress; Text type)
Exclude all values <= 1.5 bits and all Schwas.
Figure 2 bottom: Normalized Correlation Plain Text Context Corrected Phoneme info and vowel midpoint F1/F2 distance (semitones): Style X Autoprominence (X Speaker; Phoneme; Stress; Text type)
Exclude all values <= 1.5 bits and all Schwas.

Style versus Manner of Articulation
Figure 3 left: Normalized Correlation of Plain Text Context Corrected Phoneme info and duration: Position in the word X Manner (X Stress; Speaker; Style; TextType; Phoneme; Autoprominence (0, 1+2, 3+4); Syllable structure)
Exclude /@ s n t/ tokens and all values <= 1.5 bits.
Figure 3 right: Normalized Correlation of Plain Text Context Corrected Phoneme info and mid-phoneme Center of Gravity (semitones): Style X Manner (X Stress; Speaker; TextType; Phoneme; Autoprominence (0, 1+2, 3+4); Syllable structure)
Exclude /@ s n t/ tokens and all values <= 1.5 bits.
NOT USED: Normalized Correlation of Plain Text Context Corrected Phoneme info and mid-phoneme Intensity (dB): Style X Manner (X Stress; Speaker; TextType; Phoneme; Autoprominence (0, 1+2, 3+4); Syllable structure)
Exclude /@ s n t/ tokens and all values <= 1.5 bits.

Data from: Van Son and Pols, "Evidence for Efficiency in vowel production", ICSLP2002

Please note that we have updated our word frequency lists. Moreover, we have cleaned up the statistical code. The values displayed here might deviate from the ones used in the paper.

Distribution of lexical information over word position
Figure 1: Relation between segmental lexical information and the position in the word grouped by manner of articulation for comparison.

Style versus Prominence
Figure 2 top: Correlate lexical Phoneme info and Duration: Style X Autoprominence (X Speaker; Phoneme; Stress)
Figure 2 bottom: Correlate lexical Phoneme info and F1/F2 distance (semitones): Style X Autoprominence (X Speaker; Phoneme; Stress)

Stress versus Style
Figure 3 left: Correlate lexical Phoneme info and Duration: Stress X Style (X Speaker; Phoneme; Autoprominence)
Figure 3 right: Correlate Phoneme info and F1/F2 distance (semitones): Stress X Style (X Speaker; Phoneme; Autoprominence)

Normalized Correlation Analyses, using Vowel info corrected for Plain Text Context Distinctiveness and excluding all phonemes having a corrected Phoneme info value <= 1.5 bits

The average vowel midpoint F1/F2 values for point vowels in Citation speech

Vowel information as a function of Manner of Articulation and position in the word (excludes Schwa)

Style versus Prominence
Alternative for Figure 2 top: Normalized Correlation Plain Text Context Corrected Phoneme info and vowel duration: Style X Autoprominence (X Speaker; Phoneme; Stress; Text type)
Exclude all values <= 1.5 bits and all Schwas.
Alternative for Figure 2 bottom: Normalized Correlation Plain Text Context Corrected Phoneme info and vowel midpoint F1/F2 distance (semitones): Style X Autoprominence (X Speaker; Phoneme; Stress; Text type)
Exclude all values <= 1.5 bits and all Schwas.

Stress versus Style
Alternative Figure 3 left: Normalized Correlation Plain Text Context Corrected Phoneme info and vowel duration: Stress X Style (X Speaker; Phoneme; Autoprominence; Text type)
Exclude all values <= 1.5 bits and all Schwas.
Alternative Figure 3 right: Normalized Correlation Plain Text Context Corrected Phoneme info and vowel midpoint F1/F2 distance (semitones): Stress X Style (X Speaker; Phoneme; Autoprominence; Text type)
Exclude all values <= 1.5 bits and all Schwas.

Analyses, using Phoneme info corrected for Lemma Context Distinctiveness and excluding all phonemes having a corrected Phoneme info value <= 1 bits

Style versus Prominence
Alternative for Figure 2 top: Correlate Lemma Corrected Phoneme info and duration: Style X Autoprominence (X Speaker; Phoneme; Stress)
Exclude all values <= 1 bits and all Schwas.
Alternative for Figure 2 bottom: Correlate Lemma Corrected Phoneme info and F1/F2 distance (semitones): Style X Autoprominence (X Speaker; Phoneme; Stress)
Exclude all values <= 1 bits and all Schwas.

Stress versus Style
Alternative Figure 3 left: Correlate Lemma Corrected Phoneme info and duration: Stress X Style (X Speaker; Phoneme; Autoprominence)
Exclude all values <= 1 bits.
Alternative Figure 3 right: Correlate Lemma Corrected Phoneme info and F1/F2 distance (semitones): Stress X Style (X Speaker; Phoneme; Autoprominence)
Exclude all values <= 1 bits.