HNHacker News
TopNewBestAskShowJobs

jmvalin

373 karma · joined July 12, 2013

submissionscomments
jmvalin··on RNNoise: Learning Noise Suppression
Interesting -- and unexpected. I also wrote the Speex suppressor and one of the things that specifically annoyed me about it was the robotic noise and the pseudo-reverberation it adds to the speech, but it seems like some people (like you) like that. Trying to understand exactly what you don't like about RNNoise... is it how the remaining background noise sounds or how sharply it turns on/off?

I did a quick hack to RNNoise to smooth out the attenuation and prevent it from cancelling more than 30 dB. I'd be curious if it improves or makes things worse for you (compared to the samples in the demo): https://jmvalin.ca/misc_stuff/rnn_hack1/

jmvalin··on RNNoise: Noise Suppression with Deep Learning
The "distortion" you're hearing is actually just the leftover noise. When there's noise and speech at the same time and frequency, then you can year some of both and it sounds a bit harsh -- like distortion.

In general, this problem is tough because when there's a change in the signal, you have just 10 ms to decide whether it's the noise or the signal changing.

jmvalin··on Opus 1.2 Released
Even that should happen relatively quickly since so far we've also preserved API and ABI compatibility (since 1.0). Also, we're working closely with the Firefox and Chrome teams so they know what's going on with Opus.
jmvalin··on Opus 1.2 Released
All Opus releases since 1.0 are fully compatible with the specification (RFC 6716) and all future releases will also be. That's why we have standards. If we ever decided to make something that's not compatible, it would be called a different name.
jmvalin··on Opus 1.2 Released
The threshold for "transparency" varies a great deal across users. Some people I know cannot ABX 64 kb/s Opus from the uncompressed original, yet once in a while we hear of someone being able to ABX one particular sample all the way up 192 kb/s. There's also the issue of whether you have the reference to compare. Personally, I stop being able to ABX Opus somewhere around 128 kb/s, but I won't be able to tell that a song has been compressed with Opus at 96 kb/s unless I have the original to compare it to.
jmvalin··on Opus 1.2 Released
At low bit-rate, lame applies a low-pass and resamples to a lower sampling rate. If it didn't to that, MP3 would end up completely starved of bits (trying to code all frequencies) and have really atrocious artefacts. Based on personal experience and the results from a few listening tests I saw, I think you can assume that Opus can now give you about the same quality as MP3 for about 60% of the bits. For example, 76 kb/s Opus should be around the same quality as 128 kb/s MP3.
jmvalin··on Revisiting Daala Technology Demos
Obviously, the rates for these images are the same across all codecs. Choosing the rate for such comparison is also a little tricky. If you choose a rate that people would use to make their image look really good (which how most people compress images), then you have a realistic use case, but nobody can see much difference. If you choose too low a rate, then all images look really ugly. So I tried picking a rate where the resulting images are tolerable, but low enough that differences across codecs are easily noticeable.
jmvalin··on Should All Research Papers Be Free?
There is an alternative: you host your thesis (and/or papers) on your own website, or you put them on arXiv.org. It's called self-archiving and it's allowed by most publishers. Funny thing is: making your papers available for free also increases the number of citations, something academics really care about.
jmvalin··on Forging an Alliance for Royalty-Free Video
You have to start somewhere. Expect more companies to join.
jmvalin··on Forging an Alliance for Royalty-Free Video
Well, DRM is completely independent from the codec. You can put DRM on a free codec and you can have a non-free codec without DRM. As for Hollywood... note that the Alliance also includes Netflix.
jmvalin··on Cisco to release BSD-licensed H.264 stack
There is no release yet, but Cisco is saying they will be releasing the source code of the full implementation, not just nvidia-style hooks to a binary blob. AFAIK, the plan is for users to be able to replicate the build and see that the binary they got from Cisco indeed corresponds to the source code.
jmvalin··on New developments in the Opus audio codec
CPU usage in the decoder is currently higher than Vorbis, though most of the difference is due to having less optimizations. As for the encoder, it's already faster than Vorbis and should become even faster in the future. I haven't checked, but I suspect the same is true when compared to MP3. That being said, the complexity is pretty much negligible on a desktop machine.
← PreviousPage 2 of 2