System Requirements
ReSing is a 64-bit application and requires a 64 bit CPU and Operating System.
Mac® (64-bits)
Minimal: Apple M1 or Intel® Core i5 Processor, 8 GB of RAM, macOS® 13.7 or newer.
9 GB of hard drive space.
Requires an OpenGL 2 compatible graphics adapter.
Supported Plug-in formats (64-bit): Audio Units, VST 3, AAX.
Windows® (64-bits)
Minimal: Core i5 Processor or equivalent, 8 GB of RAM, Windows 10 (64 bit) or newer.
9 GB of hard drive space.
Requires an ASIO compatible sound card.
Requires an OpenGL 2 compatible graphics adapter.
Supported Plug-in formats (64-bit): VST 3, AAX. Requires an OpenGL 2 compatible graphics adapter.
Internet connection is required for authorization.
The Resing Modeler allow to create a fully functional ReSing Model from a set of vocal or instrument recording. This processing known as modeling involves analyzing and learning the unique features of a voice/instrument through a series of machine learning training steps, during which the software processes each recording to extract detailed acoustic information and generate a model capable of reproducing that voice/instrument with high fidelity.
Because the training phase relies on complex deep learning computations, hardware performance plays a crucial role in determining both the speed and quality of the results.
On macOS, Intel-based systems represent the least efficient option for voice modeling, as the process relies exclusively on CPU computation. Since traditional CPUs are not optimized for the parallel processing required by deep learning, training performance is significantly limited.
In contrast, Apple Silicon architectures (M1, M2, M3, etc.) deliver a major improvement thanks to Metal Performance Shaders (MPS), which allow the Modeler to harness the GPU and Neural Engine for accelerated processing.
On Windows, CPU-only configurations face similar limitations to those seen on Intel-based Macs, however, when equipped with an NVIDIA GPU* supporting CUDA*, Windows systems achieve the best overall performance. CUDA acceleration enables massive parallelization, allowing the Modeler to train even complex, high-quality voice models with exceptional speed and stability.
| Platform | Architecture / Hardware | Acceleration Type | Approx. Training Time | Example System | Notes |
|---|---|---|---|---|---|
| Windows | NVIDIA GPU with CUDA | GPU (CUDA cores) | ~20 min (10 min dataset) / ~4 h (1 h dataset, high quality) | RTX 5090, 24 GB VRAM | Recommended configuration for fastest training |
| Windows | CPU | CPU only | >24 hours (10 min dataset) | Intel i9 / AMD Ryzen 9 | Limited by lack of GPU parallelization |
| macOS | Apple Silicon (M1/M2/M3) | MPS (GPU + Neural Engine) | ~3 hours (10 min dataset, low quality) / >24 hours (1 h dataset, high quality) | M3, 32 GB RAM | Highly optimized through Metal Performance Shaders |
| macOS | Intel | CPU only | ~24 hours (10 min dataset) | Intel i9, 32 GB RAM | Least efficient configuration — no hardware acceleration |
*NVIDIA GPU Compute capability 5.0–12.0
*CUDA versions 11.8, 12.6 and 12.8
Drivers version ≥ 572.61
For a detailed mapping of compute capability to specific GPU models, see Nvidia’s official compute capability table here.
Frequently Asked Questions
Yes. ReSing supports ARA2 (Audio Random Access), a technology that allows seamless integration with compatible DAWs, enabling you to process audio directly on the track without manual file transfers.
Currently, ReSing supports ARA mode in the following DAWs:
- Cubase 12.0 or later
- Logic Pro X 10.4 or later (Intel architecture only)
- Studio One 4.0 or later
- Nuendo 11.0 or later
- Reaper 5.97 or later
Absolutely. ReSing can also be used as a standard plugin in any DAW. Simply load ReSing on a track, import an audio file via the built-in file browser or by dragging it into the drop area, process it using your preferred settings, and then drag the processed file from ReSing back into your DAW track. If you want to tweak the effect, just reprocess the file and replace it in the same way.
You can create your own model using the ReSing Modeler. To build a model, you’ll need about 15 minutes of clean audio from the source you want to model (a vocal or an instrument). Depending on your system’s performance, the modeling process can take as little as 20 minutes on a powerful NVIDIA RTX GPU, or several days if running on a CPU-only system.
To achieve the best results with ReSing Modeler, it’s essential to record a clean, consistent, and expressive dataset, the collection of audio files used to train your AI voice/instrument model. The model will replicate every nuance, strength, and flaw present in your recordings, so quality and consistency are key.
- Recording Content
For an high quality model record around 60 minutes of expressive singing in a single vocal style and language. Include:- Sustained vowels, pitch runs, and common phrases that cover the full vocal range (low, mid, high).
- Natural inflections, vibrato, and phrasing that represent your intended style.
- Technical Specifications
- Format: 24-bit / 48 kHz, mono, uncompressed (.wav, .aiff, or .flac)
- Microphone: Professional large-diaphragm condenser
- Interface: Clean, transparent preamp or audio interface without coloration
- Recording Environment
Use an acoustically treated or dampened space to avoid reflections and background noise. Keep the microphone away from walls or reflective surfaces, and monitor the input through headphones while recording.
The singer should wear closed-back headphones to prevent sound leakage into the mic.
Each ReSing model is trained in a specific language, which is listed among the model’s characteristics. If you process a track sung in a different language, the result may include artifacts or unnatural pronunciations, since the model isn’t familiar with those phonetic sounds. For best results, use a model that matches the language of the original performance.
When creating a model in the ReSing Modeler, you’ll be asked to specify the language used in your dataset. This ensures the model is trained correctly based on the phonetic structure of that language. If your language isn’t available and you select another one instead, the resulting model may produce artifacts or pronunciation errors, as it won’t include the unique sounds of that language.
Yes. ReSing models can also be used for spoken voice, but some specific models — labeled Narrator — are designed especially for voiceover, podcasting, and narration, as they are trained specifically on spoken performances.
The Unlimited Model Generation add-on allows ReSing and ReSing MAX owners to create an unlimited number of voice models using the ReSing Modeler.
It’s the perfect upgrade for users who want complete freedom to experiment and build as many custom voices as they wish.
Requires a licensed copy of ReSing or ReSing MAX to use.
A ReSing Session gives you full access to a selected Session Singers Showcase model for one month. It is a one-time purchase with no automatic renewal, allowing you to download, use, and release music with that voice model during the active period.
A ReSing Session can be activated on one device at a time.
