TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

DeepFilterNet: Noise supression using deep filtering

217 点作者 nitinreddy88将近 2 年前

11 条评论

orbital-decay将近 2 年前
Frankly, what I hear is very similar to the results of classic spectral denoising, even with the characteristic FFT artifacts (for Linux, there&#x27;s Noise Repellent [1] available for advanced spectral denoising; there&#x27;s also a ton of commercial spectral processors available).<p>The demonstration could use more random background noises to separate it from FFT noise suppressors (as it&#x27;s the primary benefit of ML-based filters), and more varied speech to separate it from RNNoise [2] which tends to suppress breath and cut the sibilants in an unnatural manner. The latency is also important - is it as low as in RNNoise? What about the CPU load?<p>[1] <a href="https:&#x2F;&#x2F;github.com&#x2F;lucianodato&#x2F;noise-repellent">https:&#x2F;&#x2F;github.com&#x2F;lucianodato&#x2F;noise-repellent</a><p>[2] <a href="https:&#x2F;&#x2F;github.com&#x2F;werman&#x2F;noise-suppression-for-voice">https:&#x2F;&#x2F;github.com&#x2F;werman&#x2F;noise-suppression-for-voice</a>
评论 #36227119 未加载
nitinreddy88将近 2 年前
Integrate with Pipewire: <a href="https:&#x2F;&#x2F;github.com&#x2F;Rikorose&#x2F;DeepFilterNet&#x2F;blob&#x2F;main&#x2F;ladspa&#x2F;README.md">https:&#x2F;&#x2F;github.com&#x2F;Rikorose&#x2F;DeepFilterNet&#x2F;blob&#x2F;main&#x2F;ladspa&#x2F;R...</a><p>Youtube demo: <a href="https:&#x2F;&#x2F;youtu.be&#x2F;EO7n96YwnyE" rel="nofollow">https:&#x2F;&#x2F;youtu.be&#x2F;EO7n96YwnyE</a><p>Paper Explanation: <a href="https:&#x2F;&#x2F;youtu.be&#x2F;it90gBqkY6k" rel="nofollow">https:&#x2F;&#x2F;youtu.be&#x2F;it90gBqkY6k</a>
rektide将近 2 年前
It&#x27;s so excellent how many moats are just getting obliterated.<p>I absolutely have been a real snarky hater against AI, as a horrible fuedal unobservable black box that has way too much power in the world. But open source has been doing amazing at reading the papers &amp; reproducing &amp; it&#x27;s glorious to see.<p>Amazing examples of a peership culture in action. Rising each other up is so divine. Share the knowledge &amp; means.
评论 #36221892 未加载
评论 #36223709 未加载
评论 #36222057 未加载
WiSaGaN将近 2 年前
It looks like the library in Rust is using `tract-onnx` to do the inference: <a href="https:&#x2F;&#x2F;github.com&#x2F;Rikorose&#x2F;DeepFilterNet&#x2F;blob&#x2F;2a84d2a1750a5fcb608323d1b4f964d9f1c037a6&#x2F;libDF&#x2F;Cargo.toml#L112">https:&#x2F;&#x2F;github.com&#x2F;Rikorose&#x2F;DeepFilterNet&#x2F;blob&#x2F;2a84d2a1750a5...</a> I am wondering whether using Python for research, training in big data center, and Rust at edge for efficient inference would be a trend in the future. We do have a larger community of C++ right now for inference (e.g. ggml). But Rust crate as component to build applications of AI is joy to use.
评论 #36224322 未加载
narrationbox将近 2 年前
Since it does the signal processing in the Fourier domain, does this suffer from audio artefacts e.g. hissing in the output? Torch&#x27;s inverse STFT uses Griffin-Lim which is probabilistic and if you don&#x27;t train it sufficiently, you may sometimes get noise in the output.<p><a href="https:&#x2F;&#x2F;pytorch.org&#x2F;docs&#x2F;stable&#x2F;generated&#x2F;torch.istft.html#torch-istft" rel="nofollow">https:&#x2F;&#x2F;pytorch.org&#x2F;docs&#x2F;stable&#x2F;generated&#x2F;torch.istft.html#t...</a><p>An alternative would be to use a vocoder network (or just target a neural speech codec like SoundStream).
评论 #36223266 未加载
ZoomZoomZoom将近 2 年前
The demo with the vac is certainly not a success.<p>I sometimes wonder if all those filters optimise for a wrong thing. Removing noise is meaningless, unless the overall discernability improves. If you remove noise with the price of the voice becoming choppy, &quot;robotic&quot; and unnatural, you didn&#x27;t improve the situation, and in some cases you can say only made it worse.<p>What even further deteriorates legibility for most noise suppression filters is the discrepancy between the completely dry pauses and the remaining ambiance &quot;under&quot; the voice. It would be much more interesting to see some style transfer for voice ambience as an alternative to current de-verbs.<p>When dealing with voice processing I advocate for restraining from noise suppression filters for as long as possible, and I haven&#x27;t seen a publicly available noise suppression filter which could change my position yet.
评论 #36224907 未加载
sniglom将近 2 年前
Sorry if hijacking,<p>I have a drone (DJI FPV) with a microphone. It can pick up some sounds, but the loudness from the rotors makes it really hard to hear in playback.<p>The rotor noise varies in frequency and has several harmonics as well, so it can&#x27;t be band passed.<p>I understand that you can&#x27;t get a clean or great signal from it, but something would be nice.<p>What tool would be good to use to filter out that noise?
评论 #36278211 未加载
评论 #36226401 未加载
stonelazy将近 2 年前
One of the challenges we face in this research problem is the lack of reliable metric to evaluate the quality of the NN model. In recent times, i came to know of [3Quest metric](<a href="https:&#x2F;&#x2F;cdn.head-acoustics.com&#x2F;fileadmin&#x2F;data&#x2F;global&#x2F;Datasheets&#x2F;Analysis_Software&#x2F;ACQUA_Options&#x2F;ACOPT-21-35-3QUEST-3QUEST-SWB-FB-ACQUA-Option-6844-6866-Data-Sheet.pdf" rel="nofollow">https:&#x2F;&#x2F;cdn.head-acoustics.com&#x2F;fileadmin&#x2F;data&#x2F;global&#x2F;Datashe...</a>) being helpful in this regard. Anybody have any experience with this metric ? May be in comparison with Microsoft&#x27;s DNSMOS ?
stan_kirdey将近 2 年前
<a href="https:&#x2F;&#x2F;github.com&#x2F;haoheliu&#x2F;voicefixer">https:&#x2F;&#x2F;github.com&#x2F;haoheliu&#x2F;voicefixer</a> is also a nice CLI tool to do general speech restoration<p>Demo page: <a href="https:&#x2F;&#x2F;haoheliu.github.io&#x2F;demopage-voicefixer&#x2F;" rel="nofollow">https:&#x2F;&#x2F;haoheliu.github.io&#x2F;demopage-voicefixer&#x2F;</a>
boneitis将近 2 年前
If you (especially on behalf of any hip, popular platforms like Discord) undertake any projects to aggressively denoise or compress audio, please (PLEASE) do us people with auditory processing difficulties a favor, and include such people in your testing.<p>I beg of you, with utmost sincerity.
评论 #36226645 未加载
chpatrick将近 2 年前
How does it compare to rnnoise?
评论 #36224633 未加载