[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: voiced/unvoiced detection

Pierre Divenyi <pdivenyi@MARVA4.NCSC.MED.VA.GOV> wrote:
>That is, if I were able to
>whisper louder than the highway noise, I could be perfectly intelligible
>speaking in a car with the windows down, despite the fact that the
>autocorrelation of the speech I am producing would be absolutely flat.

Not absolutely flat.  If the excitation is white noise, the autocorrelation
function (of the waveform) should look like the impulse response of the
vocal tract.  No period peaks to be sure, but not flat either!  Another way
to put it is to say that ACF and squared magnitude of the vocal tract
transfer function are Fourier Transform pairs.  Features in one translate
to dual features in the other.

For whispered speech, one should probably distinguish the issues of
transmitting segmental information ("phoneme" identity, etc.), and
intonation.  To the extent that segmental information is carried by
spectral shape, it is coded equally well if the excitation is noise-like.
A speech recognizer trained on voiced speech should work on whispered
speech. In principle.  In practice there are issues such as the different
spectral slopes of voiced and whispered excitation, and the fact that
speakers might not articulate the same when they whisper as when they use

Intonation is another problem, as it is usually thought of as being coded
by F0 which is absent in whispered speech.  I think it has been suggested
that F1 might be used in the place of F0 (how to reconcile this role with
that of coding segmental information is another mystery).  Other parameters
are timing and intensity.  Introspection tells me that whispered
articulation is more marked than voiced articulation, something akin to a
sort of "Lombard effect".  It may be a mistake to equate "whispered speech"
with "voiced speech minus the F0".


Email to AUDITORY should now be sent to AUDITORY@lists.mcgill.ca
LISTSERV commands should be sent to listserv@lists.mcgill.ca
Information is available on the WEB at http://www.mcgill.ca/cc/listserv