50% off Lifetime · code SAMPLESUMMER

Sample rate explained: 44.1 vs 48 vs 96 kHz for hardware samplers

Technical

Sample rate numbers appear in every audio spec sheet: 44.1 kHz here, 48 kHz there, sometimes 96. Getting the wrong one on your hardware usually means audio that plays back at the wrong pitch. This article covers what sample rate actually means, why two standards coexist and won’t consolidate, what happens when you load the wrong one onto your hardware, when 96 kHz is worth the extra file size, and how to think about conversion quality.

What sample rate means

Digital audio is a stream of amplitude measurements taken at regular intervals. The sample rate is how many measurements per second: 44,100 times per second for 44.1 kHz, 48,000 for 48 kHz.

The sampling theorem (usually called Nyquist, after Harry Nyquist, one of its contributors) gives you the highest frequency you can represent accurately: exactly half the sample rate. At 44.1 kHz, you can represent frequencies up to 22.05 kHz. At 48 kHz, up to 24 kHz. At 96 kHz, up to 48 kHz.

Human hearing tops out around 20 kHz, depending on age and the individual. So 44.1 kHz gives just enough headroom to capture the full range of what most people can hear. 48 kHz adds a bit more headroom above that ceiling. 96 kHz captures frequencies well past any biological relevance.

This is also why audio playback hardware includes low-pass filters before the digital-to-analog conversion step. Without filtering, content at or above the Nyquist limit produces aliasing: high-frequency content that folds back into the audible range as distortion or odd tonal artifacts.

Why two standards exist

The 44.1 kHz standard came from the CD era. When the format was standardized in the early 1980s, 44.1 kHz was chosen partly for engineering reasons related to early recording methods that used video tape as storage. It became the consumer audio standard and stuck.

The 48 kHz standard came from professional broadcast. TV and film production settled on 48 kHz, and professional gear (DAT machines, mixing consoles, most professional DAWs) followed. Today, 48 kHz is the default for video production and most professional audio work.

The result: two rates that coexist, used for different contexts, with no clean mathematical relationship between them. Converting 44.1 kHz to 48 kHz isn’t a simple integer multiple operation. The conversion ratio is 160:147, which requires real computation to do without introducing artifacts. Hardware samplers have to pick one, and most do: they expect files at a specific rate and handle everything else poorly or not at all.

What happens when the rate doesn’t match

Most hardware samplers don’t inspect the sample rate stored in a WAV file’s header before playing back. They play the raw sample data as-is, assuming it was recorded at their native rate. Load a file at the wrong rate and the hardware plays all the samples at its own clock speed.

The Make Noise Morphagene runs at 48 kHz. If you load a 44.1 kHz file, it plays those samples at 48,000 per second instead of the correct 44,100. That’s 8.84% faster, a pitch shift of about 0.88 semitones sharp. The reel sounds recognizable but is slightly faster and higher than it should be. This is the most common cause of “something sounds off” on the Morphagene.

The ALM Busy Circuits Squid Salmple runs at 44.1 kHz. Feeding it a 48 kHz file has the opposite effect: the module plays those samples at 44,100 per second instead of 48,000, so the audio comes out about 8.84% slow, roughly 0.88 semitones flat. For textural material you might not notice; for anything pitched, it’s immediately wrong.

The Elektron Octatrack takes a stricter position: it simply refuses to load WAV files that aren’t 44.1 kHz. No silent mismatch, no wrong-pitch playback, just a file that won’t appear in the browser. Some hardware enforces the rate; most doesn’t and just plays at the wrong speed.

The pitch shift from a rate mismatch is always the same calculation: (loaded rate / correct rate) − 1, expressed as a percentage speed difference, then converted to cents. Going from 44.1 to 48 kHz gives you +8.84% speed, which is about +146 cents, or roughly 1.5 semitones. Going from 48 to 44.1 kHz is −8.16%, about −139 cents. Small enough to miss on first listen with ambiguous material, obvious enough on a pitched instrument or a drum loop.

Who wants what

Here are the native sample rates for commonly used hardware, from their specifications:

  • Make Noise Morphagene: 48 kHz. Needs 32-bit float, 48 kHz, stereo WAV.
  • ALM Busy Circuits Squid Salmple: 44.1 kHz. Needs 16-bit, 44.1 kHz, mono WAV.
  • Elektron Octatrack MKII: 44.1 kHz. Refuses to load files at any other rate.
  • Intellijel Multigrain: 48 kHz. 16-bit stereo WAV.
  • Squarp Rample: 44.1 kHz. 16-bit WAV.
  • Erica Synths Sample Drum: 48 kHz. 16-bit mono WAV.
  • 1010music Bitbox mk2: 48 kHz preferred. Accepts 16/24/32-bit WAV.
  • Instruō Lúbadh: 96 kHz native. Accepts other rates with on-module conversion.

The 44.1/48 split loosely tracks the consumer/professional divide: drum-machine-influenced hardware tends toward 44.1, DSP-heavy and granular-focused hardware tends toward 48. There are exceptions on both sides, so checking the manual for any new instrument before preparing samples is worth the five minutes.

When 96 kHz matters

For most sample playback, 96 kHz doesn’t help. If you’re recording a kick drum and triggering it from a sample player, 44.1 kHz captures everything the speaker can reproduce and everything your ears can hear. Doubling the sample rate doubles the file size with no audible improvement for straight playback.

The cases where 96 kHz starts to pull its weight:

Heavy pitch-shifting or time-stretching. When audio is slowed down significantly, granular and phase vocoder algorithms are operating on data from above 22 kHz (if it exists in the source). Starting at a higher sample rate gives those algorithms more material to work with, which often produces cleaner results at extreme settings. Modules that do tape-style vari-speed, like the Lúbadh, record natively at 96 kHz partly for this reason: audio slowed to half speed is now working with content originally captured up to 48 kHz, and it sounds cleaner than starting at 44.1 kHz and hearing the artifacts at low speeds.

Aliasing-sensitive granular processing. Granular synthesis chops audio into tiny grains and manipulates them. At 44.1 kHz, certain grain sizes and overlap parameters can interact with the Nyquist limit and produce aliasing artifacts. Running at higher sample rates pushes those artifacts above the audible range. The tradeoff is memory and processing load.

Source material with heavy downstream processing. If a sample will be pitched, stretched, or used across multiple contexts, a 96 kHz source gives you more flexibility. A sample prepared today for one device might end up heavily processed somewhere else.

For most hardware playback, though: match the hardware’s native rate and don’t spend storage on 96 kHz. A clean 44.1 kHz file sounds identical to a 96 kHz file played through a hardware sampler at 44.1 kHz, because the hardware’s output stage discards everything above Nyquist anyway.

Conversion quality

When you need to convert between rates, the quality of the algorithm matters. Resampling has to either interpolate new sample points (when going up in rate) or apply a low-pass filter and decimate (when going down). Done well, the result is transparent. Done poorly, it introduces aliasing or smeared transients.

The two main failure modes:

Aliasing on downsample. If the anti-aliasing filter isn’t steep enough when reducing the sample rate, high-frequency content from the original can fold down into the audible range. This shows up as a metallic, gritty quality on transient-heavy material like percussion.

Pre-ringing on transients. Steep linear-phase filters can cause energy to appear just before a transient (pre-ringing). For drum samples this softens the perceived attack. Better resampling algorithms (minimum-phase designs, or very high filter orders with enough taps) reduce this, at the cost of more computation.

In practice, the major software tools (SoX, Audacity, ffmpeg’s libswresample, most DAW sample rate conversion engines) produce transparent results for most material. The difference between a good and a poor resampler becomes audible mainly on high-frequency test signals, extreme processing chains, or content that was already at the edge of the Nyquist limit.

The process itself is simple: set your DAW’s project rate to the target rate, import your source file, and export. The DAW handles the conversion. For large libraries, sox handles it in batch from the command line with quality flags (the --rate option with -v for very high quality uses a 148-tap filter by default).

A word on 32-bit float

Sample rate and bit depth are separate parameters, but they come up together often enough to address briefly here. 32-bit float isn’t a higher sample rate; it’s a different amplitude representation. Floating-point encoding means the signal has essentially infinite headroom during processing: no matter how many gain operations you stack, you won’t clip internally. This is why the Morphagene requires it alongside its 48 kHz rate requirement. The granular processing engine applies many amplitude operations to audio in real time, and 32-bit float gives those operations room to work without distortion. Most other hardware samplers don’t use 32-bit float because they’re playing back audio rather than processing it heavily in real time.

For the container side of audio format decisions, the WAV vs AIFF vs FLAC guide covers why hardware prefers WAV and when the other formats are worth knowing about.

The short version

Know your hardware’s native rate before you prepare samples. Prepare files at that rate. If you have a library of mixed-rate source material, convert in batch rather than file by file. Reserve 96 kHz for modules that use it natively or for source material that will be heavily processed. Good software resampling is transparent to most ears on most content.

If you’re working across multiple devices with different rate requirements and don’t want to track conversions manually, SampleStack reads WAV, AIFF, and FLAC at any source rate, converts to the exact sample rate your target hardware requires, and validates the result against the device spec before writing to your card or drive.

← Back to News