Speech Filing System

History of SFS

The past ...

The Speech Filing System (SFS) has its origins in the software developed for speech research by University College London, Imperial College London and GEC Hirst Research Centre under an Alvey information technology initiative project called SPAR.

The SPAR software was designed to support collaborative research across multiple sites. We designed a file format that would record the origin and the processing history of each data set. We standardised the format of waveforms, fundamental frequency tracks, annotations, synthesizer control data and spectrograms. We created standards for files formats, graphics, program use and program documentation, and then provided subroutine libraries to support them.

We reported on the general aims and achievements in this paper:

M.A.Huckvale, D.M.Brookes, L.T.Dworkin, M.E.Johnson, D.J.Pearce, L.Whitaker, "The SPAR Speech Filing System", European Conference on Speech Technology, Edinburgh, 1987.

When the SPAR project ended (around 1987), the ownership of the software was rather complicated. At UCL we wanted to continue to develop it and to share our developments with others who were not part of the original project. Although we felt we could give away the programs we had developed ourselves, what about the infrastructure: in particular the subroutine libraries? To get around this, Mark Huckvale rewrote the subroutine library supporting the file format and data sets, calling the result the Speech Filing System. Now UCL could distribute the sources for all parts of all the application programs it had developed. SFS was born.

There was some, but not much, continued development at the other SPAR partner sites. In particular, work at Imperial College by Mike Brookes on a signal processing library and some graphics components were eventually integrated into the software UCL distributed. Speech work at GEC Hirst Research centre was eventually abandoned and the programs they developed have been lost.

Some of the early work was done in computer languages that are now not so easy to support: in FORTRAN and Pascal; these programs have also fallen by the wayside. Also a set of programs for speech synthesis by rule were written and unfortunately lost.

The Present ...

SFS has been in continuous use in Phonetics and Linguistics at UCL since 1987. We have used it to underpin our research work in voice, in speech perception and hearing, in speech synthesis and speech recognition. We also use it in our teaching laboratory on courses such as "Acoustics of Speech and Hearing", or "Introduction to Speech and Hearing Science", or "Speech Processing by Computer".

We continue to find SFS useful for our purposes and we feel that it may be useful to others. However, we have no resources to fund the development of SFS; we have no full-time software developers in the department, nor anyone able to provide technical support. If SFS works for you and you find it useful, then we are pleased. If it doesn't work or doesn't do what you want, then all we can say is that we are sorry; we simply cannot undertake to solve other people's problems.

Recent developments in SFS have been directed at making it more friendly to PCs running flavours of Windows. The introduction of the SFSWin control program is a big step in this direction. SFSwin converts a pile of programs with command-line interfaces into a kind of windowed application. The same could be done on Unix using TCL/TK if someone had the time.

Also new is a simple stand-alone program for beginners called SFS/WASP. WASP is a Windows application that records and displays speech signals, spectrograms and fundamental frequency tracks. It saves its files in SFS format so that they are accessible to the main SFS package.

The Future ...

I have always expected that someone would develop some software that would take over from SFS. When Entropic launched ESPS (Entropic Signal Processing System) and it was take up by all the main speech research laboratories (including us, by the way), I thought the era of SFS was at an end. However when I tried to use ESPS, I found it really unfriendly, and I never put in the effort required to get familiar with its operation. It was undoubtedly flexible and of good quality, but it was not suitable for beginners. It was also very expensive (because it was a supported commercial product) and only ran on engineering workstations. Now that Entropic itself has ceased trading, there is no obvious alternative. SFS is not a professionally designed system like ESPS, but it does what we want at UCL. It is free, comes with all its program sources and runs on Windows PCs. If you like it, why not contribute new and better programs compatible with the SFS format?


Although I take much responsibility for the current state of SFS (good and bad), many people have contributed to the design, tools and code. In particular I would like to thank:

  • Mike Brookes (Imperial College London)
  • the late Mike Johnson (formerly UCL)
  • David Pearce, Lynn Whitaker (formerly GEC)
  • Andrew Simpson (formerly UCL)
  • Andrew Breen (formerly UCL)
and the many users of SFS who have sent in suggestions and bug reports.

Mark Huckvale

© 2000 Mark Huckvale University College London