Lossless Audio Passthrough in Video Watermark Removal

1. The Generational Loss Problem

Audio compression algorithms (such as AAC, MP3, and Opus) are lossy perceptual codecs. When audio is decoded to PCM and re-compressed through an encoder, cumulative quantization noise is introduced, causing:

• High-frequency attenuation above 16 kHz (loss of air, cymbal shimmer, and transient snap).
• Phase smear across stereo imaging and collapse of 5.1 surround sound metadata.
• Dynamic range compression and introduction of pre-echo artifacts on vocal transients.

2. Container Architecture: Separating Video from Audio

An MP4 or MOV file is not a raw stream; it is an ISO Base Media File Format container (ISOBMFF) composed of nested structured atoms (moov, trak, mdia, mdat). The video samples (H.264 / HEVC / AV1) and audio samples (AAC / Opus / PCM) exist in distinct, non-interfering tracks.

3. The Lossless Passthrough Pipeline

AURA ERASE preserves master audio quality using strict direct stream copying:

1. Demuxing: In the browser (via mp4box.js or WebCodecs) or on the native engine (via FFmpeg), the audio track is extracted as an untouched byte buffer.
2. Video Filtering: Only the visual video frames undergo watermark deblending and GPU acceleration.
3. Remuxing: The untouched audio track is packet-copied directly into the final container using -c:a copy (or browser mp4-muxer).

Because the audio stream never passes through a lossy codec, 100% of bit-depth, sample rate (48kHz/96kHz), and channel layout (Stereo, 5.1, Dolby Atmos) are preserved down to the exact bit.