@audio/decode-tta

Decode TTA (True Audio) lossless audio to PCM samples, in pure JavaScript

npm install @audio/decode-tta

import decode, { decoder } from '@audio/decode-tta'

TTA1 is Alexander Djourik's lossless codec: per channel, an adaptive two-state Rice coder feeds an 8-tap sign-sign-LMS "hybrid" filter and a fixed first-order predictor; multichannel streams decorrelate the last channel against the rest with a cascading half-sum/difference. Written from the TTA project and the BSD-licensed reference decoder — Djourik & Zhilin, True Audio Software (2004): ttadec.c / filter.h / ttadec.h / ttalib.h, via Rockbox's libtta port which retains the original license — see LICENSE.ttadec. Decodes exactly what ffmpeg -c:a tta writes: 8/16/24-bit PCM, mono through 16 channels.

let { channelData, sampleRate } = decode(ttaBytes)

let dec = decoder()
let head = dec.decode(chunk1)   // synchronous; header/seek-table/frame boundaries carry over
let tail = dec.flush()
dec.free()

API

decode(src: Uint8Array | ArrayBuffer): AudioData

Decode a complete stream.

decoder(): TTADecoder

Synchronous streaming decoder. decode(chunk) returns the samples of every complete frame in the data so far — frame byte lengths are known from the seek table, so a chunk boundary anywhere (even mid-header) just carries over to the next call. flush() drops a trailing incomplete header/seek-table/frame. free() releases state and is idempotent. errors counts frames dropped for a CRC32 mismatch.

AudioData

{ channelData: Float32Array[], sampleRate: number }

CRC32

TTA1 checksums the header, the seek table, and every frame independently. A header or seek-table mismatch throws — those describe the stream's shape and can't be decoded around. A frame mismatch is not fatal: since each frame resets its filter/predictor/Rice state from scratch, one bad frame can't corrupt its neighbors, so that frame's samples are dropped, errors increments, and decoding continues.

Notes

• Only format 1 (plain PCM) is decoded; format 2 (password-encrypted TTA) throws.
• Channel order is kept exactly as encoded — TTA does not reorder channels the way AC-3/DTS do, so channelData[i] is stream channel i.
• A leading ID3v2 tag (some encoders other than ffmpeg prepend one) is detected and skipped before the TTA1 signature; a trailing APEv2 tag is ignored.
• A truncated file loses its final incomplete frame silently, matching flush()'s "drop the partial" behavior in the other atoms in this umbrella.
• 60 s of 44.1 kHz stereo decodes in ~225 ms on a current laptop (~265× real-time).

Use when: opening .tta files, or decoding a TTA track pulled out of a container.

Part of @audio/decode — the decode family umbrella.

MIT © audiojs, algorithm ported from the BSD-licensed TTA reference decoder — see LICENSE.ttadec