What is the SD file format?
.sd files are binary speech and audio data files in the ESPS (Entropic Signal Processing System) format, developed by Entropic Research Laboratories. ESPS and the accompanying WAVES+ graphical waveform tool were widely used in computational linguistics and speech recognition research throughout the 1980s and 1990s.
Each .sd file begins with a 4-byte magic number (00 00 77 A0 in big-endian) followed by an extensive header - documented in the research community as unusually long and complex - that records sample rate, channel count, sample type (16-bit integer, 32-bit float, or 64-bit double), byte order, and arbitrary user-defined parameter fields. Raw audio samples follow immediately after the header.
Entropic Research Laboratories was acquired by Microsoft in 1999, after which WAVES+ and ESPS were discontinued. Speech corpora archived in .sd format are now found primarily in legacy academic datasets; conversion to WAV or NIST Sphere (.sph) is required for use with current tools. SoX and Awave Studio are among the few applications that can still process ESPS .sd files.
The .sd extension is shared by unrelated formats (such as Rational Rose sequence diagrams), so format identification should rely on the magic number rather than the extension alone.
Security & safety
RISK: LOWBinary audio data files with a complex header - not executable. Safe to open with Awave Studio or SoX. The complex header structure means malformed files could theoretically trigger parser bugs in old ESPS tools.
Format details
in a nutshell- Sound Designer I audio file (Digidesign) - Digidesign Sound Designer format (.sd or .sd1) is an early Mac audio format used by Pro Tools precursors; should NOT be confused with ESPS SD.
- Rational Rose Sequence Diagram - IBM Rational Rose UML tools use .sd for sequence diagram files; unrelated to audio.
Programs that open SD files
Technical details
deep spec| Magic number | `00 00 77 A0` at offset 0 (big-endian); appears as `A0 77 00 00` on little-endian platforms |
| Byte order | Variable - both big-endian and little-endian supported; actual byte order declared in the file header |
| Header structure | Extensive binary/ASCII hybrid header (typically several kilobytes) preceding audio data; noted in documentation as unusually long |
| Sample types | 16-bit short integer (PCM), 32-bit float, or 64-bit double - sample type declared in the header |
| Typical sample rates | 8 kHz, 16 kHz, or 22.05 kHz for speech research; exact rate stored in the header |
| Channel count | Declared in header; monaural most common for speech research recordings |
| Typical file size | 100 KB to several hundred MB for full speech research recordings |
| User-defined parameters | Header supports arbitrary key-value metadata fields beyond the standard fields - a design feature of the ESPS container |
| Original platform | Unix/Linux (ESPS was a Unix toolkit); also accessible on Windows via Awave Studio and legacy tools |
| Tool compatibility | Readable by Awave Studio, SoX (Sound eXchange), and HTK (Hidden Markov Model Toolkit, Cambridge) |
| Extension conflict | `.sd` is also used by Rational Rose sequence diagrams and other unrelated formats; the magic number distinguishes ESPS files |
| Successor format | Speech research standardized on PCM WAV or NIST Sphere (`.sph`) after ESPS discontinuation; Entropic Esignal (`.esig`) was the intended internal successor |
| Released | circa 1985-1988 (Entropic ESPS toolkit; exact release date unconfirmed) |
| Latest version | WAVES+ 5.x / ESPS 5.3 era (pre-1999); development stopped after Microsoft acquisition |
| Specification | www-mmsp.ece.mcgill.ca |
SD conversions
Community Q&A
asked by usersNo questions yet - be the first to ask about SD files.