꾸리꾸리 우잉뿌
'ㅑㅇ
DeepFilterNet GitHub Repository
- Lightweight deep filtering framework for real-time full-band (48kHz) speech noise suppression.
- Supports both Rust and Python with command-line tools and LADSPA plugins for real-time microphone noise suppression.
- Pretrained models available; works on CPU and GPU, suitable for embedded environments such as hearing aids.
Cycle-Consitency-Audio-Noise-Filter GitHub Repository
- An open-source audio filtering project applying a modified CycleGAN architecture for speech noise removal.
- Supports frequency and waveform transformations and can learn from asymmetric datasets.
- Simple and effective for various noise conditions, ideal for research and rapid prototyping.
HiFi-GAN Hugging Face Model
- High-performance GAN vocoder model converting mel spectrograms into high-quality speech signals.
- Offers dramatically faster inference speed compared to WaveNet, optimized for real-time speech synthesis and enhancement.
- Uses a generator combined with multi-scale and multi-period discriminators for simultaneous model compression and speech quality.
- Widely used for speech restoration, text-to-speech, and noise reduction enhancement applications.