Roland SP-404SX sample format

format.md

Roland SP-404SX sample file format

Notes

Field types are marked using C-style notation:

• char[4] indicates a 4-byte fixed-width string
• uint[6] indicates a 6-byte sequence
• uint8, uint16, and uint32 are unsigned byte, short, and long integers
• int8, int16, and int32 are signed byte, short, and long integers
• long double is an 80-bit IEEE 754 floating-point number.

Fields that vary are marked in bold. All other fields should be constant across samples.

File name

• Sample files are stored in SP-404SX/ROLAND/SP-404SX/SMPL/
• Sample names are in the form A00000001.WAV, i.e. one letter for the bank followed by seven digits, padded with leading zeros, for the pad number.
• Samples can be either WAVE (.WAV) or AIFF (.AIF).

AIFF sample file format

This format is generated by the sampler when recording samples to the SD card.

Samples are 44.1 kHz 16-bit PCM files in AIFF format with an additional Roland-specific header. They can be either mono or stereo.

The sample is an AIFF file with the structure:

• AIFF chunk
  • COMM chunk
  • RLND chunk
  • SSND chunk

This is a big-endian format, i.e. the most significant byte comes first and bytes appear in the order written. For example, 176,400 is 2B110h and appears in the file as 00 02 B1 10.

AIFF chunk

Field	Type	Bytes	Data	Notes
ChunkID	char[4]	4	FORM
ChunkSize	uint32	4	00 0D BF 78	see below
Format	char[4]	4	AIFF

Notes:

• ChunkSize is the length of the file remaining, i.e. the total file length in bytes minus eight.

COMM chunk

Field	Type	Bytes	Data	Notes
ChunkID	char[4]	4	fmt
ChunkSize	uint32	4	00 00 00 12
NumChannels	uint16	2	00 02	1 or 2
NumSamples	uint32	4	00 03 6F 60	see below
BitsPerSample	uint16	2	00 10	16-bit
SampleRate	long double	10	40 0E AC 44 00 00 00 00 00 00	44100 Hz

Notes:

• Chunk size is the number of bytes remaining in the chunk. It is always 18.
• NumChannels is 1 for mono, 2 for stereo.
• NumSamples is the number of samples in the audio data. For 16-bit stereo audio, this will be 1/4 of the length of the audio data in bytes.

APPL chunk

Field	Type	Bytes	Data	Notes
ChunkID	char[4]	4	APPL
ChunkSize	uint32	4	00 00 01 C2
Signature	char[4]	4	RLND
Device	char[8]	8	roifspsx
Unknown	uint8[3]	6	00 00 00
SampleIndex	uint8	1	04	see below
Padding	uint8[]	434	00 00 ...	all zeros

Notes:

• SampleIndex is always 4 for samples recorded on the device.
• This chunk is sized and padded with zeros to ensure that the the sample data starts exactly at offset 512.

SSND chunk

Field	Type	Bytes	Data	Notes
ChunkID	char[4]	4	SSND
ChunkSize	uint32	4	00 0D BD 88	see below
Offset	uint32	4	00 00 00 00	always 0
BlockSize	uint32	4	00 00 00 00	always 0
Data	int16[]	?

Notes:

• ChunkSize is equivalent to the number of samples ×
• Data always starts at offset 512 in the file.
• Data is a sequence of samples. Each sample is a pair of left and right values for a stereo file.
• ChunkSize is 8 + NumSampleFrames × NumChannels × BitsPerSample / 8

WAVE sample file format

This format is generated by the Roland Wave Converter application.

Samples are 44.1 kHz 16-bit PCM files in RIFF WAVE format with an additional Roland-specific header. They can be either mono or stereo.

The sample is a RIFF file with the structure:

• RIFF chunk
  • fmt chunk
  • RLND chunk
  • data chunk

This is a little-endian format, i.e. the least significant byte comes first and bytes appear in reverse order. For example, 176,400 is 2B110h and appears in the file as 10 B1 02 00.

RIFF chunk

Field	Type	Bytes	Data	Notes
ChunkID	char[4]	4	RIFF
ChunkSize	uint32	4	20 37 02 00	see below
Format	char[4]	4	WAVE

Notes:

• ChunkSize is the length of the file remaining, i.e. the total file length in bytes minus eight.

fmt chunk

Field	Type	Bytes	Data	Notes
ChunkID	char[4]	4	fmt
ChunkSize	uint32	4	12 00 00 00	18 bytes
AudioFormat	uint16	2	01 00	PCM
NumChannels	uint16	2	02 00	1 or 2
SampleRate	uint32	4	44 AC 00 00	44100 Hz
ByteRate	uint32	4	00 1B 02 00	see below
BlockAlign	uint16	2	04 00	see below
BitsPerSample	uint16	2	10 00	16-bit
ExtraParamSize	uint16	2	00 00

Notes:

• The chunk name is f m t space.
• Chunk size is the number of bytes remaining in the chunk. It is always 18.
• NumChannels is 1 for mono, 2 for stereo.
• ByteRate is computed as SampleRate × NumChannels × BitsPerSample / 8, or more simply as SampleRate × BlockAlign.
• BlockAlign is computed as NumChannels × BitsPerSample / 8, i.e. 2 for a mono sample and 4 for stereo.

RLND chunk

Field	Type	Bytes	Data	Notes
ChunkID	char[4]	4	RLND
ChunkSize	uint32	4	CA 01 00 00
Device	char[8]	8	roifspsx
Unknown	uint8[3]	3	00 00 00
SampleIndex	uint8	1	05	see below
Padding	uint8[]	443	00 00 ...	all zeros

Notes:

• SampleIndex starts at zero for A1 and increases by 12 for each bank, i.e. A1 = 00h, A5 = 04h, B5 = 10h.
• This chunk is sized and padded with zeros to ensure that the the sample data starts exactly at offset 512.

data chunk

Field	Type	Bytes	Data	Notes
ChunkID	char[4]	4	data
ChunkSize	uint32	4		see below
Data	int16[]	?

Notes:

• ChunkSize is equivalent to the number of samples × BlockAlign
• Data always starts at offset 512 in the file.

Pad info file

This is stored alongside the samples in PADINFO.BIN and contains 120 × 32-byte records, one for each pad from A1 to J12.

In this file, values are stored in big-endian order, i.e. an offset of 512 is 200h and stored as 00 00 02 00.

Field	Type	Bytes	Data	Notes
OrigSampleStart	uint32	4	00 00 02 00
OrigSampleEnd	uint32	4	00 02 37 28
UserSampleStart	uint32	4	00 00 02 00
UserSampleEnd	uint32	4	00 02 37 28
Volume	uint8	1	7F	0-127
Lofi	uint8	1	00	0/1
Loop	uint8	1	00	0/1
Gate	uint8	1	01	0/1
Reverse	uint8	1	00	0/1
Format	uint8	1	01	see below
Channels	uint8	1	02	1 or 2
TempoMode	uint8	1	00	see below
OrigTempo	uint32	4	00 00 04 B0	see below
UserTempo	uint32	4	00 00 04 B0	see below

Notes:

• TempoMode is 0 = off, 1 = pattern, 2 = user.
• Format is 0 for an AIFF sample, and 1 for a WAVE sample. It is possible that this corresponds solely to the endianness of the data (0 = big endian, 1 = little endian).
• OrigTempo and UserTempo appear to be beats per minute mutiplied by 10, i.e. 4B0h = 1200 = 120 bpm. The Roland Wave Converter application computes the original tempo as 120 / sample length.
• Sample start and end offsets are relative to the original file: Wave Converter sets this to 512 because the length of the RIFF headers before the raw wave data is exactly 512 bytes.

References

• Audio Interchange File Format: "AIFF"
• WAVE PCM soundfile format
• Hacking the 404SX

Really helpful! Thanks for sharing!

You mention:

OrigTempo and UserTempo appear to be beats per minute mutiplied by 10, i.e. 4B0h = 1200 = 120 bpm. 
The Roland Wave Converter application computes the original tempo as 120 / sample length.

What is the definition of the sample length ? Do you have any reference for the statement above or this is something you figured out yourself?

What is the definition of the sample length ? Do you have any reference for the statement above or this is something you figured out yourself?

I figured it out myself through observation. Sample length would be the number of samples, i.e. length in bytes divided by 4 for stereo or length in bytes divided by 2 for mono.

I found that the RLND Unknown chunk was 4 bytes (or one large int) when working with own PAD_INFO.BIN file. I'm curious if they changed the header in a software revision as I couldn't get anything but the latest version to run. I took what you wrote and added it to my SP-404SX sample manager and libs, thank you so much for taking the time to document the formats!

https://github.com/MatthewCallis/super-pads
https://github.com/uttori/uttori-audio-padinfo
https://github.com/uttori/uttori-audio-wave