Loris for Your Cough

1

Loris for Your Cough Roshan Mansinghani,
Esmeralda Martinez, James McDougall, Travis
McPhail
Goal Analyze the possibility of removing
short-time noise, such as coughs or sneezes, from
live recorded audio files.

• 2.) Reassigned Bandwidth-Enhanced Method of
  Additive Synthesis Fitz
• Implements MQ method.
• Differs in handling noise as to eliminate
  introduced errors
• Short, jittery tracks can be considered noise.
• Removes short tracks while still conserving
  signal energy and frequency centers.

• Our Algorithm
• Noise is represented by short duration tracks.
• If a track had large gaps (multiple windows)
  between partials it was broken into smaller
  pieces.
• Each track was analyzed for duration.
• If a tracks duration was less than a threshold
  it was removed.
• The signal was reconstructed using standard MQ
  methods.

Motivation Often during live recorded concerts
people cough or sneeze. This noise appears in
the recording and stands out from the surrounding
music.
Partial tracks before filtering
Before and after removal of noise tracks Fitz

• Increases bandwidth of tracks in the vicinity of
  the rejected track.
• Increases bandwidth using Bandwidth-Enhanced
  Oscillators.

• Approach
• Record a simple audio file, such as a clarinet
  playing a single note with a cough in the middle.
• Break the file up into short-time windows.
• Analyze the frequency content of each window
  separately and remove unwanted noise.
• Reassemble the file with as little distortion to
  the music as possible.

Partial tracks after filtering

• Results
• The noise frequencies were completely removed
  including the low frequency components.
• The low frequency, long lived tracks were
  preserved.
• Most of the upper harmonic information was also
  lost.

Effect of Bandwidth-Enhanced Oscillator on a
single frequencyFitz

• Background
• 1.) The McAulay and Quatieri (MQ) Method
• Window off overlapping sections of the signal.
• Compute Fourier Transform of each window and find
  dominant frequencies (partials).
• Connect partials from each window to track their
  progression through time.

• Implementation Loris Sound Software
• A C library implementing the Bandwidth-Enhanced
  Model.
• Handles windowing of signal using a Kaiser window.

• Future Research
• Improved algorithm to not remove upper harmonics.
• Automated removal of noise.
• Use Loris sound morphing capabilities to morph
  two or more sound files.
• Possible removal of other types of extraneous
  noise (cell phone, keys, clapping, etc.)

Magnitude of Kaiser Window
Frequency Response of Kaiser Window

• Computes Short-time Fourier Transforms.
• Tracks the progression of Partials through time.
• Uses the reconstruction process defined in the MQ
  model.
• Graphical User Interface (Fossa) for viewing
  amplitude and frequency tracks.

• Interpolate between connected points to generate
  a smooth track.
• Use the tracks to develop cosine terms with
  time-varying amplitude, phase, and frequency.
• Re-assemble sound by summing cosine terms.

• Contact Information
• Roshan Mansinghani rosh_at_rice.edu
• Esmerelda Martinez esme_at_rice.edu
• James McDougall jamesmcd_at_rice.edu
• Travis McPhail tjice_at_rice.edu

When re-assembling noisy signals articles are
introduced into the reconstructed signal.
Frequency Track for a clarinet playing a single
note.
Acknowledgements Kelly Fitz, Lippold Haken, and
other Cerl Sound Group Members. Susanne Lefver,
developer of Fossa. Dr.Baraniuk.