Detail Technologies · on-device models

Studio sound.
Lightning fast.

Detail’s Clear models clean speech to studio quality, on device, faster than the cloud. 12 noisy recordings from our apps: boats, demo rooms, podcasts, different mics.

Model family · Clear

Two voices.
One model.

Same architecture, two target recipes. Pick the one that fits the recording: broadcast-clean, or natural with the room left in.

Clear Studio.

Studio-quiet voice, no room noise. The default in Detail. Best when you want podcast-ready output from any input.

4.5 MB · 48 kHz · fp16 · ANE
Clear Natural.

Same enhancement, but with comfort noise preserved so silences breathe. Sounds like the same room, just cleaner.

4.5 MB · 48 kHz · fp16 · ANE
Live A/B · 12 recordings

Press play.
All ears.

Two opinionated audio models. Studio gives a warm, studio-like sound. Natural keeps the room you recorded in. Slide between no processing and full enhancement.

Loading clips…

Benchmark · Clear Studio · iPhone 17 Pro

Studio sound.
Faster on device.

The Apple Neural Engine runs Clear at 81× realtime. Five minutes of audio cleans in under four seconds. And your audio never leaves the phone.

5 minutes audio, clean in seconds
iPhone 17 Pro
3.7s
iPad Pro M4
4.1s
iPhone 15 Pro
4.9s
10-minute m4a · 8 MB · using 5G
Dolby Enhance
On cloud
~98s
Clear Studio
On iPhone 17 Pro
7s
14× faster. And your audio never leaves your phone.
Training

Built on
our own corpus.

A DeepFilterNet 3 backbone, half-sized for the Neural Engine, fine-tuned on Detail and Subwave recordings. Boats, hotel rooms, demo days, podcasts. With a warm and present audio profile, not just clean.

Available now

Better content.
Faster.

Built for Detail and Subwave. Try the models in our apps: