Poetic sound and silence

« previous post | next post »

Following up on "Political sound and silence", 2/8/2016, here's a level plot of speech segment durations and immediately-following silence segment durations from William Carlos Williams' poetry reading at the Library of Congress in May of 1945:

For comparison, here's the same thing for a year's worth of weekly radio addresses from Barack Obama (2010) and from George W. Bush (2008):

Update — and Wallace Stevens:

Also Ronald Reagan:

And Sylvia Plath:

Plus W.H. Auden:

And Gertrude Stein:

Charles Bernstein:

Carl Sandburg:


John Ashbery:

Allen Ginsberg:


  1. Terry Hunt said,

    February 12, 2016 @ 10:57 am

    Could this lead to assessments of PQ (Poeticality Quotient)?

  2. Rubrick said,

    February 12, 2016 @ 3:31 pm

    so much depends

    a small white

    ringed with bright

    showing speech

  3. Adam said,

    February 13, 2016 @ 3:34 pm

    Mind posting your code? What's extracting the segments from the raw audio data (what strikes me as by far the most difficult and potentially most error-prone portion of the process).

    [(myl) There's a better speech activity detector (SAD) in a package of signal processing, visualization, and machine learning code that Neville Ryant will be releasing on github soon.

    This one, also written by Neville, is simple but works well on decent-quality audio: in my experience, it agrees with human annotators about as well as they agree with one another. It's structured as a two-state HMM, with more or less the obvious architecture, and we've been using it for a while at LDC for tasks like pre-segmenting audio for transcription purposes. There's a description of its use here, in a page of instructions for an on-going phonetics course.

    I think that Neville is reluctant to get involved in supporting this version, but if you write to me I can send you a copy.]

RSS feed for comments on this post