🇬🇧 English | 🇹🇷 Türkçe
Autonomous, Hardware-Friendly Acoustic Upmix Engine for Home Theater and Large Media Archives
Modern stereo-to-surround conversion workflows often rely on AI source separation models such as Demucs or similar neural networks. While these systems can isolate vocals and instruments surprisingly well, they frequently introduce undesirable artifacts such as synthetic ringing, phase inconsistencies, pumping, and digital smearing.
nUpmix takes a different path.
Instead of attempting to reconstruct audio with artificial intelligence, it preserves the integrity of the original recording by relying entirely on acoustic physics, deterministic phase mathematics, and proven matrix decoding principles.
The result is a transparent, studio-inspired surround expansion engine designed specifically for:
- 🎬 Home theater enthusiasts
- 📚 Massive movie archives
- 🎵 Legacy stereo and mono recordings
- 🔊 AV receivers that deserve clean, clipping-free audio
No AI hallucinations.
No synthetic reverbs.
No unnecessary processing.
Just clean mathematics.
Before processing begins, FFprobe automatically inspects the source.
nUpmix identifies whether the input is:
- Mono (1.0)
- Stereo (2.0)
The appropriate processing matrix is then selected automatically.
Unsupported or already multichannel sources are rejected, protecting both the processing pipeline and downstream AV equipment from accidental misuse.
Modern blockbusters rely on enormous dynamic range.
Quiet whispers...
...followed by explosions.
nUpmix preserves this cinematic dynamic contrast instead of crushing it with aggressive compression.
A transparent True Peak limiter (FFmpeg alimiter) is applied only when absolutely necessary, trimming peaks above −0.5 dBFS while leaving the original dynamics intact.
No pumping.
No loudness war.
Just protection against digital clipping.
Legacy video files often suffer from imperfect frame timing.
To prevent long-term audio drift, the processing chain concludes with:
aresample=48000:async=1000
This guarantees reliable synchronization throughout playback.
Output formats are standardized as:
- Dolby Digital AC3 5.1
- AAC Stereo
up to 640 kbps, ensuring excellent compatibility across virtually all home theater systems.
Video streams remain untouched using direct stream copy.
Meanwhile nUpmix automatically detects:
- Embedded subtitles
- External
.srt - External
.ass
For Turkish archives, Turkish subtitles receive automatic:
- Default flag
- Forced flag
allowing media libraries to become immediately ready for viewing without manual remuxing.
Unlike AI upmixers, nUpmix uses carefully designed FFmpeg pan matrices to distribute sound according to acoustic principles.
The original stereo channels remain 100% untouched.
Soundstage width, stereo imaging and musical body remain exactly as intended.
Dialogue is created from a carefully attenuated L + R sum.
Instead of overwhelming the mix, voices become naturally anchored to the screen while preserving the openness of the front stage.
Rear speakers receive the L − R differential signal.
This naturally suppresses centered elements such as:
- Lead vocals
- Main dialogue
- Central bass energy
Additional processing includes:
- 300 Hz High-Pass filtering
- Optional Haas delay (10–20 ms)
creating convincing cinematic spaciousness without artificial reverberation.
Instead of the conservative 80 Hz crossover commonly used elsewhere, nUpmix routes low-frequency energy through a 120 Hz Low-Pass filter, producing fuller cinematic impact for explosions, engines and large-scale effects.
Some stereo recordings contain almost identical left and right channels.
When phase correlation exceeds approximately 95%, attempting to synthesize surround channels often produces nothing but muddy ambience.
The Autonomous Telemetry Engine detects this condition automatically.
Rather than forcing artificial rear information, nUpmix intelligently switches into 3.1 Purist Mode, distributing audio only to:
- Front Left
- Front Right
- Center
- LFE
The rear channels remain silent, maximizing clarity and preserving the original recording.
Mono recordings contain no spatial information.
Creating artificial surround effects from pure mono usually results in echo, comb filtering and an unnatural "bathroom" acoustic.
nUpmix refuses to fake spatial information.
Instead, the mono signal is copied equally to:
- Front Left
- Front Right
- Center
forming a massive and stable Front Sound Wall.
Rear channels remain completely silent, preserving the honesty and focus of the original recording.
Classic films often benefit from gentle restoration.
Optional processing modules include:
DynAudNorm
Raises quiet dialogue intelligently without destroying the natural dynamics of the soundtrack.
A gentle 50 Hz High-Pass filter removes:
- electrical hum
- HVAC rumble
- infrastructure noise
while leaving dialogue essentially untouched.
Unlike AI upmix solutions, nUpmix requires no CUDA, no GPU inference, and no multi-gigabyte neural models.
Requirements:
- Python 3.x
- PyQt5
- FFmpeg
- FFprobe
Both FFmpeg and FFprobe must be available through the system PATH.
When available, NVENC hardware encoding is detected automatically for significantly faster preview rendering.
No manual configuration is required.
nUpmix follows one simple principle:
Do not recreate sound. Reveal what was already there.
Every decision in the processing chain favors:
- Transparency
- Phase coherence
- Acoustic realism
- Hardware compatibility
- Deterministic behavior
Instead of asking AI to imagine missing audio, nUpmix lets mathematics preserve the recording's original character.
nutuzar