.TTA File Format – Frames and Rice coding

True Audio

Every frame holds the same number of samples, except the last. Frames are independent: the codec’s state starts afresh, so a decoder can begin at any of them. The compressed samples of each channel are interleaved in order, one value per channel for each sample position.

Frame layout

SizeFieldMeaning
variesBit streamThe Rice-coded values, ending on a byte boundary
4CRC-32Of the frame’s bytes before this field, little-endian

Bits are read from the least significant end of each byte first.

Two-parameter Rice code

Each channel keeps two parameters, k0 and k1, each with a running sum. Both start at k = 10, with the sum set to shift_16[k].

To read one value:

  1. Count the 1 bits up to a 0 bit: this is the unary count.
  2. A count of 0 uses k0. Otherwise subtract 1 from the count and use k1.
  3. The value is (count << k) + the next k bits.
  4. When k1 was used, update sum1 and k1, then add shift_1[k0] to the value.
  5. Update sum0 and k0 with the value (in every case).
  6. Map to a signed value: 1 + ((value >> 1) ^ ((value & 1) − 1)).

Updating a parameter

sum += value − (sum >> 4)
if k > 0 and sum < shift_16[k]:  k −= 1
else if sum > shift_16[k + 1]:   k += 1

The shift tables

shift_1[n] is 1 << n for n up to 31, then 0x80000000 up to entry 39, then 0xFFFFFFFF at entry 40. shift_16[n] is shift_1[n + 4].

After the Rice code

Each signed value goes through the hybrid filter and predictor. When every channel of one sample position is done, the channels are rebuilt from their differences.