Embedded Realtime AI Audio Enhancement

For manufacturers and developers of audio hardware and software.

Products

BUSY CONSTRUCTION SITE

1000w-michelle418_painting_By_Sophie_Taeuber-Arp._heavy_constructio_502d8ae5-07fa-406f-b729-6c9f7c3a2eed_2

Noise Removal

HANCE 3.0 Noise Removal

HANCE 3.0 offers advanced Noise Removal models that reduce background noise in realtime, providing clear and uninterrupted voice communication.

Built for modern communication systems, these models deliver fast processing, efficient CPU usage, and small model sizes, making them suitable for embedded devices and software platforms.

Use cases:

  • Embedded devices with built-in microphones (e.g., headsets, earbuds, and smart devices)

  • Live streaming and broadcasting software

  • Video conferencing tools (e.g., Zoom, Microsoft Teams, Google Meet)

  • Voice-over and post-production for media and entertainment

Technical specifications:

Latency: 20-96 ms (based on model type)
Sampling rate: 48 kHz
Number of model channels: 1
Platforms: macOS, Linux, Windows
Model size: 253 KB to 1 MB (model-dependent)
CPU requirement: Runs on 1 core of ARM Cortex-A53 or higher

New Models in HANCE 3.0:

  • speech-denoise-96ms.hance - Model for noise suppression with 96 ms latency.

  • speech-denoise-32ms.hance - Model for noise suppression with 32 ms latency, offering a balance between low latency and quality.

  • speech-denoise-21ms.hance - Model for noise suppression with 20 ms latency, optimized for ultra-low latency communication.

  • speech-denoise-tiny.hance - Lightweight model for noise suppression with 30 ms latency, optimized for small model size (242 KB).

  • speech-combined-96ms.hance - Dual-function model that removes both noise and echo, enabling clear, realtime communication.

Echo Removal

HANCE 3.0 Echo Removal

HANCE 3.0 offers advanced Echo Removal models that eliminate echo and reverb in realtime, delivering clear voice communication.

Designed for modern communication, these models offer fast processing and efficient CPU usage. They integrate with devices featuring microphones or live audio software.

Use cases:

  • Wireless hardware devices with built-in microphones (e.g., headsets, earbuds, and IoT devices)

  • Video conferencing tools (e.g., Zoom, Microsoft Teams, Google Meet) to prevent disruptive echo

  • Live streaming software for content creators, podcasters, and broadcasters

  • Assistive hearing devices to deliver clear sound for users in spaces with high acoustic reflections

Technical specifications:

Latency: 20-96 ms (based on model type)
Sampling rate: 48 kHz
Number of model channels: 1
Platforms: macOS, Linux, Windows
Model size: 253 KB to 1 MB (model-dependent)
CPU requirement: Runs on 1 core of ARM Cortex-A53 or higher

New Models in HANCE 3.0:

  • speech-dereverb-96ms.hance - Model for echo and reverb removal with 96 ms latency.

  • speech-dereverb-32ms.hance - Model for echo and reverb removal with 32 ms latency, offering improved latency-to-quality performance.

  • speech-dereverb-20ms.hance - Model for echo and reverb removal with 20 ms latency, optimized for ultra-low latency communication.

  • speech-combined-96ms.hance - Model that removes both echo and noise, enabling clear, realtime communication.

  • speech-denoise-dereverb-32ms.hance - Model for combined de-noising and de-reverberation with 32 ms latency, balancing low latency with high performance.

garage reverb

1000w-michelle418_By_Sophie_Taeuber-Arp._An_empty_car_garage_--ar_1_da338aa8-3bea-45b0-b7a5-b96d69d16e80_0

Stem Separation

1000w-michelle418_By_Sophie_Taeuber-Arp._Disco_dancefloor_full_of_h_0a843b7c-94c1-40f5-a939-ec63ebf56688_1

Stem Separation

HANCE 3.0 Stem Separation

HANCE 3.0 offers advanced Stem Separation models that isolate, adjust, and remove individual instrument tracks in realtime while music plays.

With support for vocals, piano, bass, drums, and other instruments, these models power live performance, remixing, and music production software.

HANCE models integrate with mixers, speakers, and music production tools, providing precise control and realtime processing.

Use cases:

  • Live DJing and remixing, allowing performers to isolate and adjust individual stems during live sets

  • Karaoke apps and devices for on-demand vocal removal, creating instant karaoke tracks

  • Music production software for realtime stem isolation, enabling precise edits and custom remixes

  • Sound sampling tools that extract instrument stems for reuse in new productions

Technical specifications:

Latency: 70-209 ms (based on model type)
Sampling rate: 44.1 kHz
Number of model channels: 1
Platforms: macOS, Linux, Windows
Model size: Varies by model

New Models in HANCE 3.0:

  • music-stem-separation-70ms-large.hance - Model to separate music into vocals, piano, bass, drums, and other instruments with 70 ms latency.

  • music-stem-separation-209ms-large.hance - Model to separate music into vocals, piano, bass, drums, and other instruments with 209 ms latency.

  • drumset-separation.hance - Model to separate drum sets into kick, snare, toms, hihats, cymbals, and other components with 96 ms latency.

  • movie-stem-separation-209ms.hance - Model to separate movie soundtracks into dialogue, music, vocals, sfx, and other components with 209 ms latency.

Custom models

HANCE offers a professional service for companies and enterprises seeking customized audio models. Our team of AI experts develops models to meet specific technical needs, ensuring compatibility with your business environment and operational requirements.

Custom models are ideal for businesses with unique audio challenges, such as noise removal in industrial settings, echo cancellation in large conference spaces, or models that combine multiple audio enhancement functions into one efficient solution.

Use cases:

  • Noise cancellation for environments like construction sites, factory floors, and public transit hubs

  • Echo removal for conference rooms, auditoriums, and large-scale communication systems

  • Multi-function models that combine noise removal, echo removal, and other enhancements into one model

  • Custom models for audio analysis, classification, and content filtering for business applications

Our Custom Model Process:

  • Data Collection: We analyze your company’s needs and collect data to build a problem-specific training set.

  • Model Design: Our AI experts create a model architecture to solve your audio challenges.

  • Validation: We measure the model's performance to verify its requirements.

  • Deployment: The completed model can bve integrated into your companies your software, devices, or enterprise infrastructure.

Try HANCE Models:

Evaluate the performance of HANCE models with the HANCE Audio Plugin. The plugin provides a live demonstration of our audio enhancement models. Businesses can also access SDKs and documentation for smooth integration into existing systems.

For access to pre-trained models, including noise removal, echo removal, and stem separation, visit the HANCE Model Repository. These models showcase the foundational capabilities that can be customized for your business.

Interested in a Custom Model? Our custom modeling service is available exclusively to companies and enterprise clients. Contact us to discuss your business needs. Our AI specialists will work with your team to develop a model for your specific needs.

1000w-michelle418_painting_By_Sophie_Taeuber-Arp._heavy_constructio_502d8ae5-07fa-406f-b729-6c9f7c3a2eed_2

Music Removal

Echo Removal removes echo/reverb and enhances voice clarity. Made to ensure efficient digital audio communication.

It can be integrated and embedded in microphones on devices or software, and acts as a sound filter in realtime.

Use cases:

  • Wireless hardware devices with microphones.
  • Streaming software.
  • Any software or hardware product where realtime communication is essential.

Technical specifications:

Latency: 20-107 ms (depending on model size)
Sampling rate: 48 kHz
Number of model channels: 1
Platforms: macOS, Linux, Windows
Model size: 1-3 MB
CPU reference: Can run on 1 core on an ARM Cortex-A53

 

garage reverb

1000w-michelle418_By_Sophie_Taeuber-Arp._An_empty_car_garage_--ar_1_da338aa8-3bea-45b0-b7a5-b96d69d16e80_0

Stem Separation

1000w-michelle418_By_Sophie_Taeuber-Arp._Disco_dancefloor_full_of_h_0a843b7c-94c1-40f5-a939-ec63ebf56688_1

Drum Separation

Stem Separation gives you the ability to alter music while it’s playing.

Adjust the volume of vocals, drums, bass, or piano - or remove them completely.

It can be integrated and embedded in mixers, speakers, or software.

Use cases:

  • Hardware or software for remixing music.
  • Speakers, software and apps for making instant karaoke.
  • Sound sampling.

Technical specifications:

Latency: 209 milliseconds
Sampling rate: 44.1 kHz
Number of model channels: 1
Platforms: macOS, Linux, Windows
Model size: 4 MB per instrument

 

Product Page

Keep in mind these models can be altered, combined, or customised to your needs.

Noise Removal

Noise Removal removes noise and enhances voice clarity. Made to ensure efficient digital audio communication. It can be integrated and embedded in microphones on devices or software, and acts as a sound filter in real-time.

Busy construction site

1000w-michelle418_painting_By_Sophie_Taeuber-Arp._heavy_constructio_502d8ae5-07fa-406f-b729-6c9f7c3a2eed_2

Echo Removal

Room echo, or reverb, can make speech hard to interpret. Our echo removal model has won blind tests as being best in class. Integrate it in your video conferencing software for clear communication and better user experience.

Garage reverb

1000w-michelle418_By_Sophie_Taeuber-Arp._An_empty_car_garage_--ar_1_da338aa8-3bea-45b0-b7a5-b96d69d16e80_0

Stem Separation

Our AI-model that alters music while it’s playing. No pre-loading. Adjust the volume of vocals, piano, bass and drums - or remove them completely. Made for music software and hardware. With HANCE embedded your customers can enjoy the creative possibilities of instant Stem Separation.

Stem separation

1000w-michelle418_By_Sophie_Taeuber-Arp._Disco_dancefloor_full_of_h_0a843b7c-94c1-40f5-a939-ec63ebf56688_1

Try it yourself!

This demo is using our processing API, and is not running in real-time.

Get in touch

Interested and want to learn more? Book a meeting with us!

Contact us
software-hardware-1-2

For software and hardware

Snackwave lyft cray sartorial. Everyday carry 8-bit kombucha chicharrones meh, cold-pressed XOXO poutine photo booth VHS small batch af direct trade literally pork belly.

“Since integrating HANCE's API, our video conferencing software has seen a 40% reduction in support calls related to audio issues."

 

Steven Gerrard

CEO Soundly

Add an Awesome Headline

Plus, some additional information make this section look awesome.

+50 yrs

Sound Engineering and Audio Processing

You can hear the difference

Run on anything

98%

Customer Satisfaction

Effortless and reliant streaming of realtime enhancement

FAQ

Frequently Asked Questions