woodson

27 ms·

361.
by woodson 9y ago
There's the GALE data provided by LDC ( https://catalog.ldc.upenn.edu/ ): - GALE Phase 1 Arabic Broadcast News Parallel Text - Part 1 (LDC2007T24) - GALE Phase 1 Arabic Broadcast News Parallel Text - Part 2 (LDC2008T09)
362.
by woodson 9y ago
You can create phonemic transcriptions as a back-off for unknown words (at least in WFST based setups), but with things like "sequel" this won't help much.
363.
Apollo: “Android of self-driving cars” open sourced by Baidu Research(apollo.auto)
3 points by woodson 9y ago|0 comments
364.
by woodson 9y ago
A reference is just a listing of author(s), year of publication, title, journal/conference. There's no link (perhaps a DOI). There are usually no checks (other than the reviewers/editor looking over the list of references).
365.
by woodson 9y ago
Have you compared your system to state of the art results on NIST Speaker Recognition Evaluations? I'm asking because the datasets released for these evaluations every two years or so are what have been driving progress in the field fo
366.
by woodson 9y ago
Because WaveNet is very impractical in production (per sample prediction). Tacotron may be a better fit ( https://google.github.io/tacotron ).
367.
by woodson 9y ago
Tensorflow is only mentioned for NLU, no mention of it for ASR.
368.
by woodson 9y ago
I'm thinking of driving south from Hood River towards Madras; OR-26 between Government Camp and Madras is probably going to be backed up, though (or so I imagine).
369.
by woodson 9y ago
Problem is, the output would sound pretty metallic, just like LPC10.
370.
by woodson 9y ago
It also makes it hard to disambiguate cities/airports. Entering "Portland" brings up two entries named "Portland, United States" (presumably Oregon and Maine) along with "Portland, Australia", but no way o
371.
by woodson 10y ago
This is essentially linear algebra, not behavioural psychology. I believe that one shouldn't draw such broad conclusions from a 1% improvement on some evaluation dataset.
372.
by woodson 10y ago
Take a look at Google's gemmlowp low-precision GEMM library: https://github.com/google/gemmlowp (Used in tensorflow)
373.
by woodson 10y ago
Mozilla Deepspeech has been around for a while ( https://github.com/mozilla/DeepSpeech ). In fact, this code looks like it heavily copied from there (see attribution notices in comments). Their examples use much less dat
374.
by woodson 10y ago
Intel's AVX2 instructions don't crash the machine, but lead to an extreme increase in voltage and, thus, temperature. Sustained use of these instructions (say, for example, in signal processing and iterative parameter estimation a
375.
by woodson 10y ago
Kaldi is not a point-and-click solution, it's a toolkit to develop your own speech recognition system. That said, it makes it incredibly easy if you know what you're doing, as it brings all the necessary tools and even provides so
376.
by woodson 10y ago
For the evaluation in the paper, speakers are on separate channels (mono, it's a telephone conversation, after all). Generally there are solutions for separating speakers on a single channel that can work fairly well (assuming your tra
377.
by woodson 10y ago
For anyone wondering what the recordings in the HUB5 2000 eval data (the test data) sound like: https://catalog.ldc.upenn.edu/desc/addenda/LDC2002S09.wav
378.
by woodson 10y ago
I think his point is that, because of that negative comment his methods and results will never actually put to the test of whether they will be useful to the community; however, given the volume of new papers coming out it is unlikely for t
379.
by woodson 10y ago
The quality of reviews of conference papers is mostly terrible, though. I've been on both sides of it, here are some major complaints: * There is only one round of reviews, with a binary accept/reject outcome. In case of rejection
380.
by woodson 11y ago
Except for conference papers, which are even worse. There's so much "secret sauce", only few are independently reproducible. Sadly, in some research areas they are the dominant way of publishing research output.
381.
A journal for MD/RAID5(lwn.net)
1 points by woodson 11y ago|0 comments
382.
by woodson 11y ago
Somewhat disappointing that in Australia RPi Zero costs AU$19.38 (plus AU$12.95 shipping). Quite a lot more than US$5 (=AU$6.93).
383.
by woodson 11y ago
Did anyone with a dual boot configuration already try the upgrade? This article [1] suggests that, for once, you don't have to restore GRUB after the upgrade. [1] http://answers.microsoft.com/en-us/insider/wik
384.
by woodson 11y ago
I've also experienced occasional panics when entering/exiting standby, but only when a JVM instance or Virtualbox were running at the time. Otherwise standby/suspend seem to work fine (though maybe that's just a coincide
385.
The Unreasonable Effectiveness of Character-Level Language Models(nbviewer.ipython.org)
74 points by woodson 11y ago|12 comments
386.
by woodson 11y ago
At least in academic circles, statisticians working in forensic inference and statistics widely support probabilistic approaches, to give the strength of the evidence in terms of likelihood ratios (or Bayes factors) with respect to prosecut
387.
by woodson 12y ago
Look at Table 3 in the paper. Also, 16.5% error ;)
388.
by woodson 12y ago
Providing the source would be a step to improve transparency and reproducibility (the text does not provide sufficient detail for even someone working in the field to reproduce what they did so that he would arrive at the same results); how
389.
by woodson 12y ago
A lot of the methods I listed don't have anything to do with speech per se . They DO have something to do with how to use data to e.g. remove unwanted variability or achieve better class separation. What makes them old methods is th
390.
by woodson 12y ago
I agree. I find the use of the term odd and it is not commonly used in an academic context; however, I think I recall having seen it being used informally in the context of voice commands (i.e., controlling a PC or appliance) or, more gener

More ›