FAQ
Frequently Asked Questions
-
What is HANCE?
HANCE provides machine learning algorithms that enhance audio in realtime. It removes sounds you don’t want and enhances the ones you do want.
This is done by identifying the components in audio. We call this process Audio Source Separation. In 20 milliseconds it learns which frequencies are voice, noise, or reverb - and adjusts them according to what you want to hear. But why stop there?
It can also identify different instruments(stems) in a song. This gives you access to realtime stem separation. Meaning you can control the volume of the vocals, piano, bass and drums - listen to them exclusively, or remove them completely.
With a library of 5MB and model sizes ranging from 3-4MB - HANCE is the most lightweight, low cost, and compute-efficient way to enhance audio that we’ve heard of.
-
Who is HANCE for?
Our realtime capabilities allows HANCE to be used in a variety of industries:
Improve the audio in video conferencing or intercom? Create instant karaoke from any song? Adjust ambience and dialogue in a movie?
The gate is low, and there is no hard limiter to what HANCE can do. Contact us if you have an audio problem, we might be able to help.
-
What hardware can use HANCE?
Almost any device can use HANCE. From hearing aids to supercomputers.
HANCE has a library of 5MB and model sizes ranging from 3-4MB.
It works in realtime with low latency and minimal CPU-usage, making it ideal for many lightweight use cases.
Our products run on macOS, Windows, and Linux.
It’s optimized for Intel and ARM (SIMD), and we have bindings with Python.
-
What software can use HANCE?
Any software can use HANCE.
HANCE has a library of 5MB and model sizes ranging from 3-4MB. It works in realtime with low latency and minimal CPU-usage.
We offer pre-trained models but can also tailor models for your requirements.
We also support WebAssembly.
-
How do I integrate HANCE?
HANCE offers an SDK in a variety of programming languages for easy integration.
We also offer APIs for Web, Python and C.
Find our APIs here.
It’s optimized for Intel and ARM (SIMD), and we have bindings with Python.
-
Why choose HANCE?
HANCE is built with realtime processing and audio in mind, with latencies down to 20 milliseconds.
It’s not built on top of a third party framework.
We offer a lightweight solution, with a library around 5MB and model sizes ranging from 3-4MB.
We can be implemented in both software and hardware.
HANCE is a compute-efficient way to offer realtime noise removal, voice boost, and stem separation.
-
What is stem separation?
Stem separation is what we call the process that our AI does. When it hears a mix of sounds it’s able to identify the different components(stems) in that mix. It can separate voice from reverb, bass from drums, and birds chirping from construction work – then filter accordingly to your needs. HANCE can do this in realtime.
It gives you the ability to control the volume of the vocals, piano, bass and drums in a song - listen to them exclusively, or remove them completely.
This can be done in realtime on a hardware device, or any audio file.
Make every song a karaoke song, or remove the drums and play along to your favourite band.
-
How are HANCE's AI models trained?
HANCE’s AI models are trained on a large data set of legally acquired audio.
In collaboration with one of the world's leading sound libraries, Soundly - audio engineers and machine learning experts have created a symphony called HANCE. The most compute-efficient audio enhancing engine.
-
What does it cost?
In general HANCE offers a yearly subscription based on the number of end-users of your product. Please contact us for more information or to get a customized offer.
mail@hance.ai
-
How can HANCE help me?
Speed up your development, and have AI expertise on tap.
We let you concentrate on your core strengths and reduce R&D costs.
-
Can I buy 1 license?
If you are not looking to cooperate as a business you can try HANCE as a software plugin at Acon Digital.
-
How can I test it?
Check out our APIs, or our drag-and-drop demo tester on the front page.
-
How much processing is needed?
We offer a lightweight solution, with a library around 5MB and model sizes ranging from 3-4MB.
Technical benchmark: Our smallest models can run in realtime on 1 core on a ARM Cortex-A53
-
What is HANCE working on?
Firstly, we are expanding our capabilities at removing unwanted audio.
We have made noise and reverb a thing of the past - but we’re gradually working our way through common problems like distortion, crackling, clipping, humming, buzzing, wind noise, handling noise, boom noise, microphone rustle, and more - in realtime.
Secondly, we are exploring areas such as generative AI, audio identity, deepfake verification and bandwidth extrapolation.