Audio-to-Audio Index
Speech enhancement and audio editing models, grouped
Models that clean up recorded speech: denoising, dereverberation, restoration and bandwidth extension.
Speech enhancement and audio editing models, grouped
Note Index: every model here and in Audio editing utilities, grouped by what it does.
Enhance broadcast audio quality with super‑resolution upscaling
Restore and enhance audio files with AI models
Note Found by Daniel. Lightweight denoise + dereverb, ONNX/TFLite builds for on-device use.
Note Found by Daniel. Sidon successor for dialogue: restoration with speaker separation.
Note Found by Daniel. ONNX restoration: denoise, dereverb and bandwidth extension (EN/FR).
Note Found by Daniel. NVIDIA universal speech enhancement (Mamba), any sampling rate, language-agnostic.
Note Found by Daniel. GGUF build of Sidon restoration for CPU/llama.cpp-style runtimes.
Note Found by Daniel. StuPASE: generative enhancement aiming at studio quality; removes noise and reverb, keeps speaker identity (16 kHz, cisco-open/pase).
Note Found by Daniel. Transformer voice restoration for degraded recordings (custom code).
Note Added in discovery: the original Sidon restoration model that Sidon-GGUF converts.
Note Added in discovery: current Nemotron Labs release of RE-USE (most-liked RE-USE variant).
Note Added in discovery: streaming RE-USE for real-time enhancement.
Note Added in discovery: widely used denoise + enhance model from Resemble AI.
Note Added in discovery: noise suppression + echo cancellation, GGUF, for calls and live audio.
Note Added in discovery: codec-based autoregressive speech restoration.
Note Added in discovery: real-time denoise that also suppresses background speakers, ONNX.
Note Added in discovery: SGMSE+ diffusion enhancement and dereverberation (research reference).
Note Added in discovery: classic MetricGAN+ baseline via SpeechBrain.
Note Added in discovery: MP-SENet trained on DNS challenge data, strong denoiser.
Note Added in discovery: MossFormer2 speech enhancement, MLX port for Apple silicon.
Note Added in discovery: DeepFilterNet3 (real-time noise suppression) as ONNX; upstream lives on GitHub.
Monthly audio-to-audio models on the Hub, with landmark rele
Note Reference: tracks how many audio-to-audio models are on the Hub.
Note Moved from collection 'Audio restoration' (2026-11-17 tidy).
Note Moved from collection 'Audio restoration' (2026-11-17 tidy).
Use DeepFilterNet2 to denoise audio no file size limit
Note Moved from collection 'Deep Filter' (2026-11-17 tidy).