gok

228 ms·

2071.
by gok 11y ago
I'm curious how much of the accuracy gains come from only having to run the decoder when the LSTM emits a phoneme rather than for each 10 ms frame, which presumably allows the language model search to be much more aggressive.
2072.
Google voice search: faster and more accurate(googleresearch.blogspot.com)
188 points by gok 11y ago|67 comments
2073.
by gok 11y ago
The leading complaints about the "Edit Alarm" screen is kinda weird...unless "Apple’s new direction" means Apple's new direction post-2006. The new: https://unicornfree.com/wp-content/uploads&#x
2074.
by gok 11y ago
> Don't use Safari. The code is a mess and security vulnerabilities are frequent, but slow to patch. Over the last two years, Safari has had half the vulnerabilities [1] as Chrome [2] or Firefox [3]. Security updates are frequent [4
2075.
by gok 11y ago
I'll assume by "the good fight" you meant "[making] international copyright law more permissive." Surely you see how that was in your employer's best interests for other parts of their business long before they
2076.
by gok 11y ago
In addition to XPC, there is also Mach IPC that uses the Mach Interface Generator. The best documentation for that is Mac OS X Internals: A Systems Approach, by Amit Singh, particularly Chapter 9. Unfortunately not updated since 2006, so mi
2077.
by gok 11y ago
Android's speech recognizer has a compact/offline mode, but that's definitely not what's run by default.
2078.
by gok 11y ago
There are several tremendous advantages to server-based speech recognition. Firstly, the models (particularly the language models) needed for state of the art performance are huge. It's not atypical for papers to discuss using a billio
2079.
by gok 11y ago
Both .NET Native and this (bitcode) are an AOT compilation toolchain.
2080.
by gok 12y ago
A memory unsafe version is possible in Swift too. In fact, an implementation of one is included at the top of the article.
2081.
by gok 12y ago
I don't mean to belittle you or Google's speech team, but neither homophones nor proper names are considered hard problems in modern automatic speech recognition.
2082.
by gok 12y ago
That was an even weirder comparison. They compare a system trained on 2000 hours of acoustic data mismatched with the testing data to their system, which was trained on 300 hours of matched data in addition to the 2000 hours of mismatched
2083.
by gok 12y ago
So with 300 hours of training data it does worse on SWB than a DNN-HMM, or even a GMM-HMM system? But when they give it 2300 hours or training data, it can beat those 300 hour trained systems? This is still very cool, but that comparison do
2084.
by gok 12y ago
Super cool demo. Linear programming in C++ with a Python wrapper has been implemented by many other packages too. Does anyone know how Glop compares with scipy.optimize, PuLP, pl_solve, COIN-OR's many projects, or CPLEX and its many Py
2085.
Path of certification flight of 747-8 spells out a message(flightaware.com)
4 points by gok 15y ago|1 comments
2086.
by gok 15y ago
You sure can. ^(int x, int y){printf("%d\n",x+y);}(2,4); Automatic memory management is interesting, but orthogonal.
2087.
by gok 15y ago
Is that really that much better than C blocks? #include <stdio.h> int main (int argc, char const *argv[]) { int (^add)(int, int) = ^(int x, int y) { return x + y; }; int (^max)(int, int)
2088.
by gok 16y ago
> if we codenamed Django 1.3 "Wordpress" Can I just say that I, for one, would find that awesome? :)