Inaugural addresses: SAD.

« previous post | next post »

A few days ago, I posted some f0-difference dipole plots to visualize the contrast between Barack Obama's syllable-level pitch dynamics and Donald Trump's ("Tunes, political and geographical", 2/2/2017):

Obama 2009 Inaugural Address Trump 2017 Inaugural Address

For another take on the same contrast in political prosody, I ran a "Speech Activity Detector" (SAD) on the recordings of the same two speeches, and used the results to create density plots of the relationship between speech-segment durations and immediately following silence-segment durations:

Obama 2009 Inaugural Address Trump 2017 Inaugural Address

Although Obama's modal speech-segment duration is shorter than Trump's, their mean speech-segment durations are similar, as are their overall percentages of speech-segment time. But their mean number of words per speech segment and their speaking rates are quite different:

% spch Av. spch dur Words per spch seg WPM (spch) WPM (all)
Obama2009  68%  2.4 sec.  7.72  192  131
 Trump2017  65%  2.3 sec.  5.31  137  89

This is consistent with the observations on overall speaking rate in "Slow-talking the Inaugural", 1/21/2017. Here are the same numbers for all the inaugural addresses since Kennedy:

% spch Av. spch dur Words per spch seg WPM (spch) WPM (all)
Kennedy1961 63.1 1.8 4.73 155 98
Johnson1965 47.6 1.6 3.72 137 65
Nixon1969 66.6 2.1 6.63 186 124
Nixon1973 68.9 2.2 5.91 164 113
Carter1977 52.7 1.2 3.35 162 85
Reagan1981 68.4 2.2 6.50 181 124
Reagan1985 71.3 2.3 8.85 232 165
Bush1989 61.1 2.0 6.25 188 115
Clinton1993 65.6 2.0 5.81 176 115
Clinton1997 58.3 1.8 4.98 170 99
Bush2001 62.0 1.9 5.80 180 111
Bush2005 61.6 2.2 5.99 160 99
Obama2009 68.1 2.4 7.72 192 131
Obama2013 60.3 2.4 7.67 189 114
Trump2017 65.3 2.3 5.31 137 89

(Warning: This is all based on unreliable texts from The American Presidency Project  and automated analysis of variable-quality audio from the videos therein linked, and should be checked before inappropriate levels of belief are awarded to it. For example, the text given for Lyndon Johnson's 1965 address includes an interesting passage of 761 words that was apparently omitted from the address as delivered — at least it is not to be found in the linked video of the address. I caught and corrected that particular problem, since the omitted portion is more than half of the original text, and thus resulted in implausible values for words per segment and speaking rate. But it's  likely that there are some other less egregious issues of a similar sort. In addition, the automated SAD often treat applause as "speech", though this should not cause any major changes in the statistics.)

For some background on the visualizations presented here, see "More on pitch and time intervals in speech", 10/15/2006; "Political sound and silence", 2/8/2016; "Poetic sound and silence", 2/12/2016; "Some visualizations of prosody", 10/23/2016; "Carl Kasell: diabolus in musica?", 11/5/2016.

Here is the overall speech-silence distribution combining all of the Inaugurals from Kennedy to Trump:

This continues to suggest something that's hinted at in the Obama and Trump plots, namely that there are three (or more) axes of relationship between speech-segment duration and following silence-segment duration. These might represent different levels of prosodic phrasing and/or rhetorical organization.

For completeness, here are the plots for all of the inaugurals back to Kennedy 1965:

Trump 2017:

Obama 2009 Obama 2013

 

Bush 2001 Bush 2005

 

Clinton 1993 Clinton 1997

 

Bush 1989:

 

Reagan 1981 Reagan 1985

Carter 1977:

 

Nixon 1969 Nixon 1973

Johnson 1965:

Kennedy 1961:



Comments are closed.