CB Sounddesign
Batch WAV conversion
Operator's manual · v1.0
Operator's manual · v1.0
CB Sample Maker
Batch conversion for hardware samplers — the machines with strict WAV headers, 8.3 file names, small memory and opinions about bit depth. Drop a folder of modern files in, and out comes a set the hardware will actually load, optionally with a keymap and a preset written alongside.
Target
Rate, depth, channels, resampler
Processing
Trim, normalise, fade, character
Export
Naming, folders, keymap, ZIP
Nothing leaves this machine
Every file is decoded, processed and re-encoded in the browser. There is no upload, and no file is written anywhere until you export the ZIP.
Every file is decoded, processed and re-encoded in the browser. There is no upload, and no file is written anywhere until you export the ZIP.
Chapter one
Source and profile
Device profile
A preset for a target machine, setting rate, depth, naming and header rules in one go. Start here: the point of a profile is that it knows the constraints you would otherwise discover by failing to load a file.
Drop audio files
The source list. Folders are read where the browser allows it. The table shows each file's source format, length, output name and resulting size, so the whole batch is auditable before it runs.
Preview: Source / Converted
A/B of one file before and after. The only honest way to judge the character settings — a decimated, saturated sample sounds fine in isolation and wrong next to its original.
Save this profile as JSON
The whole configuration as a file. Worth doing for every machine you own; the settings are fiddly and the second batch should not need the same thinking as the first.
Chapter two
Target format
What the files must become. These are the settings the hardware dictates rather than ones you choose freely.
Rate / Bits
Output rate and depth, including 32-bit float for the cases where the destination is software rather than hardware. Most vintage samplers want 44.1 kHz or lower and 16 bit.
Channels
Stereo, or one of the mono reductions: sum, left only, right only, mid, side, or split into two files. Mid and side are the interesting ones — mid gives a mono fold that keeps the centre, side gives the difference alone, which is rarely useful and occasionally exactly right.
Sum above correlation
Sums to mono only when the two channels are similar enough to sum cleanly, and falls back to your chosen alternative when they are not. This is the setting that prevents phase cancellation from quietly destroying a few files in a large batch.
Otherwise use
The fallback: louder channel, left, right, or mid anyway. Louder channel is the safe default.
Clean resampler
Anti-aliased. What you want almost always.
Vintage resampler
No filter, aliasing included. Not a defect — it is the sound of the machines these files are destined for, and choosing it here gives a predictable version of that character rather than whatever the hardware does.
Strict WAV header
Writes only the fmt and data chunks. Old samplers frequently refuse files with extra metadata chunks, and this is the fix for a file that plays everywhere except on the device.
TPDF dither
Applied when quantising to 16 bit or below. Leave it on — the alternative is quantisation distortion, which is worse than the noise.
Chapter three
Processing
Optional. The mode switch at the top of this block decides whether any of it runs at all.
Straight — change format only
Format conversion and nothing else. The right mode when the files are already edited and you only need them in a different container.
Full — run the processing chain
Everything below, in order. For raw material that has not been through an editor.
Remove DC offset
Removes any constant offset. Costs nothing and occasionally recovers a surprising amount of headroom.
Trim silence / Threshold / Keep pre-roll
Cuts leading and trailing silence below the threshold, keeping a few milliseconds before the start. Keep some pre-roll: trimming exactly to the first sample above threshold clips the attack.
Normalize peaks / Peak target
Raises level to a ceiling. Per file makes every sample equally loud, which destroys the dynamics of a velocity set. Match group raises the whole set by one amount, preserving the relationships. For a sampled instrument, match group is nearly always correct.
Fade in / out ms
Short fades at both ends. A millisecond or two removes the click that trimming creates.
Gain dB
A flat offset, applied after normalisation. For pulling a whole set down to leave headroom in the sampler.
Snap to zero crossing
Moves the start and end to the nearest zero crossing. The other half of the anti-click measures, and free.
Chapter four
Character
Deliberate degradation. Everything here removes quality on purpose, and everything here is audible.
Decimate
Sample-rate reduction without filtering — aliasing included. The classic lo-fi sampler sound. Judge it in the A/B preview, not by the number.
Saturation
None, soft clip, or tape-ish. Soft clip is a limiter with character; tape-ish adds low-order harmonics and softens transients.
Drive dB
How hard the saturator is pushed. Push it far enough and the trim and normalise settings upstream stop behaving as you expect — the level entering the saturator changes its character, not just its loudness.
Chapter five
Key mapping
How the files are told what pitch they are. This is what turns a folder of samples into an instrument.
Don't set a root note
Leave it out. For one-shots, percussion and anything unpitched.
Note in the source filename
Reads the note from the name. The reliable option when the recorder named the files — which it usually did.
Pitch detection
Analyses the audio. For material that arrived with unhelpful names. Check the results: detection is good on sustained pitched material and poor on percussive or noisy sounds.
Chromatic, ascending
Assigns notes in order from a start note. For sets that are known to be in order and simply unlabelled.
One fixed note
Everything on the same note. For layered one-shots that will be triggered by pad rather than by pitch.
Written as / Middle C
Whether sharps appear as Cs3 or C#3, and which octave number is MIDI 60. The # character is illegal in file names on some machines, which is why the s form exists.
Write loop / root-note chunk
Embeds the root note in the WAV itself, so a sampler that reads the chunk maps the file automatically. Turn it off alongside the strict header for the oldest hardware.
Write .sfz / .dspreset / .csv
A keymap, a DecentSampler preset, or a plain file list. Cheap to generate and worth taking even if the destination is hardware — the CSV is an inventory, and the .dspreset gives you a software version of the same instrument for free.
Chapter six
Naming, folders and export
The part that vintage hardware is most particular about.
Filename template
The pattern for output names, with the index, the note and the source name available as fields.
Start at / Digits
The first index and how many digits it is padded to. Pad to the width the machine expects, or its own sort order will differ from yours.
Pads / bank
How many files per bank, which drives the folder split when you are exporting per bank.
Case / Max name length
Upper, lower or unchanged, and a length cap. Uppercase and eight characters is the combination the oldest machines want.
ASCII only, no spaces
Strips anything else. Not optional on hardware that reads a FAT filesystem directly.
8.3 short names
Full DOS-era naming. The last resort, and the one that makes a file load on a machine that rejects everything else.
Flat ZIP / One folder per bank / Single named folder
The archive layout. Per bank matches how the hardware organises memory; a single named folder is right for software destinations.
ZIP name
The archive. Everything — audio, keymap, preset, CSV — comes out in this one file.
Run one file first
Load a single representative file, set everything, export, and load it on the target machine. A failed batch of four hundred files is the same amount of work to diagnose as a failed batch of one, and considerably more to redo.
Load a single representative file, set everything, export, and load it on the target machine. A failed batch of four hundred files is the same amount of work to diagnose as a failed batch of one, and considerably more to redo.
CB Sounddesign · Sample MakerOperator's manual · v1.0