New Applications of the Sound Description Interchange Format
Résumé
The Sound Description Interchange Format (SDIF) is a recently-adopted
standard that can store a variety of sound representations: time domain,
spectral, higher-level models, etc. SDIF's design strikes a balance between
an overly strong standard, which would restrict innovative and creative
uses, and an overly weak standard, in which it is so easy to modify the
format that each institution has its own incompatible version of SDIF, even
when using the same sound representation. SDIF's structure as a sequence of
time-tagged IFF-style chunks or "frames" supports this balanced
extensibility: extra information can be added to standard frame types, and
entirely new representations can be defined by new frame types. This paper
reviews early uses of SDIF.
IUA, IRCAM, and CNMAT have added SDIF support to existing tools for sound
analysis and synthesis, using standard frame types for STFT results,
spectral peaks, tracked sinusoids (both pseudo-harmonic or not), and
fundamental frequencies. Some applications augment these basic
representations with additional information such as perceptual salience of
each partial and time points for attack, sustain, and release for sounds
that fit the model of a single note. An important feature of SDIF is that
this kind of custom extra information can be added to standard frame types
in a way that allows programs that don't know about the extra information to
read and manipulate the basic data.
We also review experimental frame types that are being used for innovative
new sound representations not yet part of the SDIF standard. IUA's method of
taking a sinusoids plus residual model and further analyzing the residual
results in an SDIF file with multiple independent streams of data tied
together semantically into a single sound object. CNMAT's prototype model
for pipe organ sounds is based on the standard exponentially-decaying frame
type, but with additional information to give the parameters of a
cosine-shaped attack for each partial. IRCAM's Chant software uses FOF
synthesis and includes a number of new SDIF frame types.
Additional experiments involve using SDIF to generate new data structures
for the representation of spectral data. One particularly promising approach
rearranges the time tagged frames so they form a matrix of frames indexed by
pitch, amplitude, and other global properties of each spectrum.
As new uses of SDIF reach maturity they will be added to the standard;
therefore it's important for all users of SDIF to make their experimental
work known to the SDIF community. This presentation will be of interest
both to users wanting to take advantage of these sound representations and
to developers interested in integrating their work into the SDIF effort.