Audio format

This is the way the audio data is structured on the eMMC flash memory of the SP-1. There is no filesystem, simply a list of songs with title and artist names and a memory offset and length expressed in sectors.

• 24 bit 48kHz audio
• 8192 byte sectors containing 340 audio frames each
• Album metadata is stored at the start of the first sector and is padded with X's (0x58)
• Bytes 2040-2043 of each sector are reserved for timing data
• Bytes 2044-2047 of each sector are reserver for LED data

Audio data structure

The bytes we read the audio from.

All songs are 24bit, 48 kHz, 8 channel audio (4 stereo tracks)

---

Song

A song is split in sectors

---

Sector

size: 8192 bytes

A sector contains

• 340 audio frames = 7.083ms of audio at 48 kHz
• 2 bytes of timing/sync data (bytes 2040, 2041)
• 2 bytes encoding tempo (bytes 2042, 2043)
• 4 bytes of LED data (bytes 2044-2047)

A sector is split up in 4 x 2048 byte blocks

---

Block

size: 2048 bytes

A block contains

• 85 audio frames

---

Audio Frame

size: 24 bytes

An audio frame contains

• 4 stereo stem frames

Audio frames are not ordered sequentially inside a sector, but split up along the 4 blocks of a sector like so:
(blocks are ordered 0, 2, 1, 3)

Block   Frame
0       0
        4
        8
        ..
        336
2       2
        6
        10
        ..
        338
1       1
        5
        9
        ..
        337
3       3
        7
        11
        ..
        339

If we were to iterate throught the 340 frames of a sector it would go something like this:

#define FRAME_SIZE 24           // 8 * 3 bytes for 8 channel 24 bit audio
#define BLOCK_SIZE 2048         // size of a block in bytes
#define FRAMES_PER_SECTOR 340   // number of frames in one 8192 byte sector

int block_order[] = {0, 2, 1, 3};

for(int frame = 0; frame < FRAMES_PER_SECTOR; frame ++){
    int block_id     = block_order[frame % 4]; 
    int block_offset = BLOCK_SIZE * block_id ;
    int frame_offset = FRAME_SIZE * frame / 4;
    int byte_offset  = block_offset + frame_offset;
}

Where

• frame is the index of the frame in the 340 samples of a sector.
• block_id is the index of the block that contains the frame.
• block_offset is the byte offset of the block in the sector.
• frame_offset is the byte offset of the frame in the block.

The audio frame samples are structured like this:

offset  length  
0       6       stem frame 0
6       6       stem frame 1 
12      6       stem frame 2
18      6       stem frame 3

---

Stem frame

size: 6 bytes

A stem frame contains

• 2 interleaved left and right samples

In a stem frame the bytes are interleaved unconventionally: (LE 301, 452)

offset  sample  byte
0       L       MB
1       L       MSB
2       R       MSB
3       L       LSB
4       R       LSB
5       R       MB

So grabbing a stereo sample from the bytes in a frame would look something like this:

int32_t sample_left =  (data[1] << 24) | (data[0] << 16) | (data[3] << 8); // 103
int32_t sample_right = (data[2] << 24) | (data[5] << 16) | (data[4] << 8); // 254