Twitter | Search | |
This is the legacy version of twitter.com. We will be shutting it down on 15 December 2020. Please switch to a supported browser or device. You can see a list of supported browsers in our Help Center.
Loreto Parisi
MSc Computer Engineering 2006 Working on Machine Learning. Tweets are my own
13,309
Tweets
895
Following
1,100
Followers
Tweets
Loreto Parisi 7h
Conv-TasNet: Surpassing Ideal Time-Frequency Magnitude Masking for Separation 🐍
Reply Retweet Like
Loreto Parisi 8h
Pi-hole is a sinkhole software that protects every devices on the same local network from unwanted content through a device like a
Reply Retweet Like
Loreto Parisi Nov 23
How to prevent very common memory leaks in using WeakMap
Reply Retweet Like
Loreto Parisi Nov 18
These are the addon to accelerate on based processors: numpy, grpcio, h5py, cp38, tensorflow addons plus ml compute
Reply Retweet Like
Loreto Parisi Nov 18
As predicted... for with nativr acceleration supported on Macs with chip and -based Macs
Reply Retweet Like
Loreto Parisi Nov 14
I bet on support on ARM soon maybe has some insight
Reply Retweet Like
Loreto Parisi Nov 12
Replying to @migueldeicaza
You have forgotten the most resilient one
Reply Retweet Like
Loreto Parisi Nov 10
Interesting fact about new Apple chip: “Utilize ML frameworks like TensorFlow or Create ML, now accelerated by the M1 chip”. But what about support to ? Is something missing here or a new SDK is coming out?
Reply Retweet Like
Loreto Parisi Nov 7
Hey experts and engineers how would you evaluate the output quality of a adapter like this little one (aukey br-c1, CSR chip SNR >75dB)?
Reply Retweet Like
Loreto Parisi Nov 6
Is this the world we created? What did we do it for...
Reply Retweet Like
Loreto Parisi Nov 4
there’s a bit of truth here. Google gave Colab notebooks with 12 hours of free gpu running on your data. They gave research . Facebook supported torch and gave birth to . They do products but they supported research with frameworks
Reply Retweet Like
Loreto Parisi Nov 1
Replying to @faroit @keunwoochoi
Also I find that PEAQ and PESQ are missing when evaluating the quality of the audio and specifically of the vocals separated from the mixture, even if they are using MUSHRA method as for ITU Recommendation BS.1534-3
Reply Retweet Like
Loreto Parisi Nov 1
Replying to @faroit @keunwoochoi
Some highlights from the paper: the vocal separation SDR of frequency domain models decreases as the loudness of the input mixture decrease; these models are more stable from the viewpoint of the noise; A loudness normalization is used for the robustness against the loudness.
Reply Retweet Like
Loreto Parisi Oct 29
The 💵 Billion Parameter club of
Reply Retweet Like
Loreto Parisi Oct 27
is the company who produced the first large scale and commercially viable the XC2064 in 1985 - Ross Freeman was awarded with the National Inventors Hall of Fame for his invention. Good move by after latest acquire
Reply Retweet Like
Loreto Parisi Oct 27
Replying to @julien_c
aka the "unfamous fork. I have to say I was pro io.js because the" promise "(the idea not the scheme 🤪) was good. But the reality was different. Several successful" fork stories " there have been, but most some had pitfalls and relaunches. Think of WebKit, Chromium, etc.
Reply Retweet Like
Loreto Parisi Oct 26
So a implementations for Theano Ops could bring back the last static Graph based framework to life compiling it to a modern computation library!
Reply Retweet Like
Loreto Parisi Oct 26
I have published a generic inference pipeline of wav2vec 2.0 using
Reply Retweet Like
Loreto Parisi Oct 25
It could be interesting to fine tune the COLA pretrained embedding on the MTT (50 tags like guitar, classical, slow, techno, strings, etc. and MSD (50 tags like rock, pop, alternative, indie, electronic, etc.) audio tags datasets like the good musicnn does.
Reply Retweet Like
Loreto Parisi Oct 25
How to develop high-performance object detection/recognition applications for based edge devices
Reply Retweet Like