object-based audio capture
an ideal example
This first example features two sounds, five seconds long,
played simultaneously in an empty room with very little
reverberation (dB60 = 0.1 seconds). Sound A is male speech:
"but to the infuration of scientists, for no known reason, not
all of them did." Sound B is female speech: "shell shock caused
by shrapnel is sometimes cured through group therapy." The
sounds were recorded by microphones 1 and 2.
Proceed as follows:
- Listen to the two microphone signals by clicking on the
red dots in the image below.
- These two microphone signals were fed into a BSSD
algorithm, which successfully extracted the two original
sound sources from the mixtures.
- To hear the extracted sounds, click on "A" to hear the
male sample, and "B" for the female sample.
- When listening to these samples, instead of focusing on
the desired sound source, notice the attenuation of the
other source.
A real-world example
In this second example, I recorded two people speaking
simultaneously in a very noisy and reveberant conference room.
In the image below, the female speaker is at position "A" and
the male speaker is at position "B." Proceed in the same manner
as in the previous example: listen to the microphone signals
first, and the extracted sounds second.
- Upon listening to the microphone signals, observe how
reverberant the signals are.
- In output A, the male speech has not been attenuated that
much, but it has been blurred to the point of making it much
less intelligible.
- The BSSD algorithm does a fairly decent job of extracting
the male voice (output "B") from the mixture.
