Portable REMASTRA AI Audio Remaster Studio 1.5.5
Audio remastering software for music, dubbing, voice and video soundtracks. AI Denoise · 8, 53 or specialized AI Stems · AI Remaster · WAV/FLAC export from mono to Atmos 9.1.6, DTS:X & Binaural
Workflow
| Step | What it does |
|---|---|
| ① Import & Analysis | Opens WAV, FLAC, MP3, OGG, OPUS, M4A, AAC, AIFF, WMA, MP4, MOV and MKV. Measures loudness (EBU R128 / ITU-R BS.1770-4), 4× true peak, LRA, crest factor, stereo correlation and the spectrum. Detects whether the content is voice or music. |
| ② AI Denoise | Demucs voice isolation: removes music, ambience and crowd noise behind the voice, with an adjustable background level. Restoration tools: rumble filter, 50/60 Hz de-hum (8 harmonics), de-click, de-esser. |
| ③ AI Stems | Four models. Demucs v4: 8 fast stems (vocals, drums, bass, guitar, piano, strings, pads, synths). MVSep Mega BS-RoFormer: up to 53 instruments (lead vocal, backing vocals, kick, snare, hi-hat, toms, violin, viola, cello, trumpet, trombone, saxophone, flute, organ, harp, accordion, sitar…). BS-RoFormer — Male/female vocals (aufr33): splits the voice into male and female stems + instrumental residual. BS-RoFormer-1296 (viperx): high-precision vocals/instrumental split (SDR 12.96 dB). Each stem has mute/solo, a gain control and its own export, and the “Play the stem mix” button sits right at the top of the page next to “Separate stems”. |
| ④ AI Remaster | Analysis and automatic choice among 10 profiles, linear-phase corrective EQ (8k-tap FIR), adaptive multiband compression, harmonic exciter, analog saturation, M/S stereo imaging with mono bass, glue compression, loudness normalization and a true-peak limiter. Every decision the engine makes is listed. Reference-track mastering is also available. |
| ⑤ Export | WAV: 16/24-bit PCM or 32-bit float, WAVE_FORMAT_EXTENSIBLE with a channel mask, automatic RF64 above 4 GB. FLAC: 16/24-bit. Channel layouts: Mono, Stereo, 2.0, 2.1, 5.1, 7.1, 7.1.6, 9.1.6, DTS:X, Binaural. Spatialization uses the AI stems as objects, or a spectral direct/ambience upmix; Binaural folds that same 3D scene down to 2 headphone channels with ITD/ILD, spectral and decorrelation cues (not measured-HRTF convolution). TPDF dither and SoX VHQ resampling (44.1–192 kHz). |
Instant A/B listening (Original / Denoised / Master / Stem mix): click the buttons at the top of the window. The space bar starts and pauses playback.
Release Notes:
Two new BS-RoFormer stem models
BS-RoFormer — Male / female vocals (by aufr33, ≈ 330 MB): splits the voice into a male stem and a female stem, plus a computed instrumental residual — the three stems add up exactly to the mix. Great for duets, mixed choirs and features.
BS-RoFormer-1296 (by viperx, ≈ 610 MB, SDR 12.96 dB): a benchmark vocals/instrumental separation model widely used in the UVR community — an excellent default choice for isolating a lead vocal.
Both models are one-click downloads from the Stems page, cached locally like the 53-stem model, and run on lighter 8 s / 4 s / 2 s memory-mode segments (4–8 GB of VRAM, or CPU for short clips).
The model selector on the Stems page now lists 4 engines: Demucs v4 (8 stems), MVSep Mega BS-RoFormer (53 stems), and the two new specialized vocal models — each with its own memory-mode options, download button and description.
Mute/solo, gain and per-stem export now work identically across all four models; 5.1–9.1.6 object spatialization routes the new models’ vocal stems to the center channel automatically.
