The SF2 format was a miracle of 90s programming. Unlike a simple sample dump, an SF2 file contained a complete virtual instrument: the raw audio samples, a voice-stealing algorithm, low-pass filters, LFO routings, and a multi-stage envelope generator. But the TS-10’s magic wasn’t in the raw waves—it was in the behavior : the way a flute sound would morph into a choir if you held the key down, the way aftertouch added not just vibrato but a subtle distortion, the way the “Funk” wave in the Transwave section would cycle through eight different attack transients depending on velocity.
He connected the TS-10’s main outs to the Pinnacle’s inputs. He disabled the noisy internal fan on his PC. At 3 AM, with the studio dark, he began. He loaded the TS-10’s legendary preset, “DreamPad” —a cavernous, evolving swell that used two Transwaves, one reversing, filtered through a resonant low-pass. He triggered a middle C, let it sustain for 47 seconds, and hit record. He did this for every note from C-2 to C-8. He did this for the "Stereo Grand Piano," the "Warm Strings," the "ResoBass." He filled a 4GB hard drive with raw, 16-bit, 44.1kHz stereo WAVs.
Leo Focht is 73 now. He builds model ships and has perfect hearing for his age. He does not own a computer. But once a year, his grandson brings a laptop over. The grandson, a music producer named Leo III, loads up a DAW and pulls up a file. It’s always the same file. He plays a middle C. The "DreamPad" swells, its noisy, imperfect loop cycling forever, the ghost of the TS-10 breathing through a 26-year-old SoundFont.
And for a moment, 1998 and 2026 are the same year. Ensoniq TS-10 SoundFont -SF2-
In the winter of 1998, the air in the Los Angeles recording studio The Vault smelled of ozone, stale coffee, and ambition. Leo Focht, a 47-year-old sound designer with a hearing range that engineers swore defied physics, stared at the instrument that had consumed his last six months: an Ensoniq TS-10.
Leo smiles. “That’s it,” he whispers. “That’s the sound.”
Leo’s workstation was a beige Pentium II running Windows 98. His tools: a Turtle Beach Pinnacle sound card with a proprietary S/PDIF input, a copy of Chicken Systems Translator , and a mountain of pirated RAM. His process was monastic. The SF2 format was a miracle of 90s programming
Leo was fired.
To the uninitiated, the TS-10 was just a 61-key workstation synth, its grey chassis unremarkable beside a bank of Moogs and Prophets. But Leo knew better. Inside that unassuming shell lived a 24-bit polyphonic aftertouch keyboard, a proprietary synthesis engine called "TS" (Transwave Synthesis), and a 16-track sequencer that had powered half the R&B hits of the late 90s. Its sound was its secret weapon—a gritty, warm, almost tactile quality. The piano had a wooden knock; the strings breathed with a noisy, imperfect vibrato; the pads bloomed like flowers in slow motion.
The SF2 format allowed for up to 27 different modulators. The TS-10 had 16 real-time controllers. Leo spent two weeks just mapping the aftertouch to filter cutoff response. On the TS-10, it was exponential—a light touch added warmth, a hard squeeze added bite. In SF2, he had to build a piecewise linear curve. He failed. Then he failed again. Finally, he wrote a custom script in an ancient version of Python that brute-force calculated 128 breakpoints. At 4 AM on a Tuesday, he played the converted patch. He pressed down on his MIDI keyboard’s aftertouch. The sound screamed . He cried. Just a little. He connected the TS-10’s main outs to the
Three months in, with 47 patches converted, a power surge fried his Pinnacle card. The hard drive with the raw samples was corrupted. He had backups of the loops, but the original multi-samples—the 2,000+ individual notes—were gone. The TS-10 was a rental. It was due back in two days.
But the internet is a digital graveyard that refuses to stay dead. In 2002, a bedroom producer in Ukraine uploaded “TS10_Legacy.sf2” to a forgotten FTP server. In 2005, a tracker forum in Sweden embedded it into a keygen. In 2011, a sample library curator on Reddit named VintageSamples_Archive found a pristine copy on a Zip disk at a flea market in Berlin.
Leo’s mission, assigned by a boutique sample library startup called Sonic Foundry , was impossible: translate the soul of the TS-10 into the cold, sterile language of the SoundFont 2.0 (.SF2) format.
The TS-10’s samples were not perfect. To save memory (the TS-10 had only 6MB of factory ROM), Ensoniq’s engineers used clever, short loops. But translating a hardware loop to an SF2 loop was a form of torture. Leo would load a sample into Sound Forge 4.0 . He’d zoom into the waveform, looking for the "zero-crossing"—the exact point where the wave’s voltage returned to nil. He’d find a 200-sample cycle that sounded seamless on the TS-10. But in the SF2, it would click. Pop. Buzz. One night, working on the "Electric Grand" loop, Leo heard it—not a click, but a ghost. A faint, repeating artifact of the original recording session Ensoniq had used back in ’96: a distant car horn, looped into eternity. He isolated it. He named the file “TS10_EGrand_GHOST.wav” and kept it as a reminder that hardware has secrets software never can.
Today, the Ensoniq TS-10 SoundFont lives in the dark corners of thousands of hard drives. You can hear it if you know where to listen. It’s the warm, unstable pad on that lo-fi hip-hop track with 2 million YouTube views. It’s the brittle piano on that indie game soundtrack that made you nostalgic for a childhood you never had. It’s the bass in that techno track that shakes the subwoofer at 3 AM in a warehouse in Detroit.