little-scale — the chip sample packs (2008–2012)

Between 2008 and 2012 he recorded a dozen sound chips onto disk, systematically, and gave the results away. As audio they are raw material; as documents they are a survey of what each chip can be made to do, because every post states the method, the coverage and the count.

The method, which is the same every time#

• Recorded from real hardware through interfaces he built, "all recorded from hardware using custom designed interfaces".
• He generated the sounds himself: "I have created all of the sounds myself on the hardware (as in, these are not sounds or samples recorded from games or demo programs)" — repeated almost verbatim across the packs. This is a provenance claim, and it is what separates these from ripping (sample-ripping).
• Recorded dry, off the chip, before the console's analogue stage. The SP0256 pack omits "the two pole filter that normally follows the audio output from the chip"; the SAA-1099's "audio output stage of the sound chip has not been filtered". His reasoning: "you can always add a digital filter via software if you so desire."
• Format: 16-bit 44.1 kHz mono WAV, plus an MP3 mirror (96 or 192 kbps); "normalised and trimmed, but not compressed nor equalised in any fashion", with fades added to stop clicks — the same discipline as his kit tutorial (s-little-scale-lsdj-kits).
• Systematic coverage, not a highlight reel: every waveform against every pitch, every algorithm against every octave.

The packs#

The SID 8580 pack, called out#

"a sample pack, featuring every C and G note from C-2 to G 6 for the main basic waveforms of a Commodore 64 (8580 SID chip). The waveforms are triangle, sawtooth, pulse and noise. The duty cycle for the pulse wave was set to approximately 50 %. The samples were recorded from C64 hardware directly using a custom designed interface." WAV 44.1 kHz mono and MP3 192 kbps.

It is the 8580 specifically — the later, cleaner revision — so it is a document of that chip's basic waveforms rather than of the 6581's (sid). What it deliberately is not: no filter, no ring modulation, no sync, no envelope shapes beyond whatever the recording captured, one duty cycle only. "Basic" is accurate.

The SP0256-AL2 pack, called out#

The General Instrument speech chip, and the pack is a complete allophone inventory: "There are a total of 59 samples. These represent all 59 sounded allophones that the SP0256-AL2 is capable of producing via its internal ROM." Named by allophone with example words — /OY/ BOY, /AY/ Sky, /EH/ End, /KK3/ Comb, /PP/ Pow, /JH/ Dodge, /NN1/ Thin, /IH/ Sit…

His framing is not speech synthesis: "The aim of this pack is not so much to form synthesised speech, but perhaps to enable a user to explore these sounds in a creative fashion." Licence: Creative Commons Attribution-Noncommercial 3.0.

The number 59 is not a coincidence — resolved. lsdj's SPEECH instrument builds its 14 word slots "from 59 allophones", and the manual's Appendix A, "Sample Kits", names the machine behind them outright: "SP0256-AL2 / General Instruments / 1981 — The SP0256-AL2 Speech Processor IC contains a programmable digital filter that can be made to model a vocal tract" (source: s-lsdj-manual). LSDj's speech and this pack document the same chip; both counts are its 59 sounded allophones.

A technique that came with a pack: SN76489 noise bass#

The Sega Master System post of 2009-07-31 explains why one of the chips behaves oddly, and it is a usable trick:

"the noise generator either gets its frequency either from a division of the master clock, or its clocked directly from the frequency generated by pulse channel three of the SN76489. In other words, it doesn't have its own frequency divider that is set by a data register. The cool thing about controlling the noise channel from pulse channel 3 is that the output noise frequency is octaves lower than the current pulse three frequency. By changing the frequency of channel three whilst setting the volume of channel four (noise), low bass lines can be written."

So on an SN76489 the noise channel doubles as a sub-bass voice, pitched by a different channel's frequency register — you spend channel 3's pitch and channel 4's volume to get one bass line. His MIDI interface gained a "bass mode" that packages the two writes; before that he had to "write notes with a velocity less than 8 on channel three for the pitch, and notes with a velocity greater than 7 on channel four to set the volume". The later "pristine" pack includes "all timed bass sounds for noise", which is this.

The licence changed, and that matters#

The 2008–09 packs went out as Creative Commons Attribution-Noncommercial 3.0. By March 2010 he had relicensed the lot: the SAA-1099 post calls it "copyright-free and license-free" and lists "my other sample packs, all of which are now deemed copyright-free and license-free" — the ukulele, SID 8580, YM2413 pitched and drums, SN76489, YM2612, khene, SP0256-AL2, POKEY noise and Atari 2600 TIA packs. By 2012 the wording is "Royalty free, copyright free, license free, restriction free."

For a wiki that has to think about sample-ripping and attribution, that is the cleanest provenance in the field: sounds generated by one person on hardware he owned, released without restriction.

Practical takeaways#

• These are the honest way to put a chip you do not own into a tracker: hardware recordings of self-generated sounds, unrestricted, and dry enough to shape.
• Record dry and before the console's output filter, then filter in software — his standing argument, and the reason the packs still sound usable.
• A systematic pack (every waveform × every pitch) is a multisample instrument waiting to happen (multisampling), not a folder of one-shots.
• On the SN76489, tie the noise channel to pulse 3 for bass; the noise channel is not only for drums.

Relevance to the wiki#

The wiki's coverage of Sega, Atari and the speech chips is thin because sources are thin. These posts are sourced, dated statements of what those chips produce, and two of them cover chips the wiki documents properly — the sid and the SP0256 behind lsdj's SPEECH.

To verify#

• The SID pack's arithmetic does not obviously close. The title says 112 samples; the text says every C and G from C-2 to G6 across four waveforms. On the YM2413 pack he states that the same C-and-G range is "16 in total", which would give 64, not 112. Neither figure is corrected in the posts.
• Whether the 8580 recordings used a filter setting, an envelope, or a fixed gate length is not stated.
• ~~The SP0256/LSDj "59 allophones" correspondence~~ — resolved 2026-08-31 from the LSDj manual's Appendix A (see above).
• No .zip was downloaded, so the packs themselves are not in raw/ and the counts are the posts' own.
• The "custom designed interfaces" are described elsewhere on the blog (the Sega Master System MIDI interface, the C64 interface); only the sample-pack posts are ingested here.

Pages touched#

chip-samples · sample-ripping · sid · lsdj · nanoloop · multisampling · recording-samples · little-scale · levels-and-headroom