audio.cpp v0.9.4 supports NVIDIA Real-time RE-USE! No Python required.
#2
by audio-cpp - opened
Thanks to the NVIDIA team for releasing Real-time RE-USE!
audio.cpp v0.9.4 now supports Real-time RE-USE for streaming speech enhancement, with incremental audio input/output and all 30 exit-layer/lookahead configurations.
A 14-second recording achieved the following streaming results (F32, exit layer 8, zero lookahead; from the initial PR, excluding any subsequent optimizations)
| Backend / hardware | Time to first audio | RTF |
|---|---|---|
| CUDA / RTX 5090 | 7.46 ms | 0.0270 |
| Vulkan / RTX 5090 | 14.94 ms | 0.0492 |
| HIP / Radeon 8060S | 24.25 ms | 0.1652 |
| Metal / Apple M4 Pro | 21.81 ms | 0.1039 |
If you find this useful, please give us a ⭐ on GitHub or a ❤️ on this Hugging Face repo!