woodson

27 ms·

331.
by woodson 9y ago
Would you be able to tell the difference on a GSM cell phone call?
332.
by woodson 9y ago
Yes, Elman also worked on recurrent neural networks applied to language/text. A lot of word2vec stuff doesn’t seem too far off when you read the articles on clustering of recurrent (hidden) layer activations.
333.
by woodson 9y ago
Check out this one: https://github.com/wiseman/py-webrtcvad If this one does not work for your application, perhaps look into simpler ones like the ones used in mobile telephone codecs or in Speex.
334.
by woodson 9y ago
Note that the book was released 2009 and the draft is from 2007. While most of the content is still very relevant for understanding the basics and challenges in building TTS systems, recent progress in DNN-based synthesis (including WaveNet
335.
by woodson 9y ago
I’m in favor of getting rid of „neural“. That would help dispense with all the unhelpful discussions about how different DNNs work compared to the human brain.
336.
by woodson 9y ago
It's not just about the amount of data, though. The speech and audio quality of Google's TTS data is likely better than the audio contained in the LJ dataset (disclaimer: I've only listened to samples contained in the latter,
337.
by woodson 9y ago
at least the dollars are insured ;-)
338.
by woodson 9y ago
That data is no good for this purpose, as it’s from a lot of different speakers and does not have speaker labels, i.e., you can’t tell which sentences were spoken by which speaker.
339.
by woodson 9y ago
Another piece of advice: Only do a Ph.D. if you get paid to do it (in addition to not having to pay tuition).
340.
by woodson 9y ago
That challenge seems to be more about speech command recognition (isolated words). They supply 1 second long recordings of 30 short words. > There are only 12 possible labels for the Test set: yes, no, up, down, left, right, on, off, sto
341.
by woodson 9y ago
With the new tax regulation, as a U.S. citizen, do you have to pay taxes on tuition and stipends you receive when you do a Ph.D. overseas? (I'd assume so, since overseas income is taxable AFAIK, but I don't really know what the si
342.
by woodson 9y ago
Whether that is “open access” enough for commercial use is an interesting question. I thought that the SCOTUS recordings, for example, can not be used for commercial applications, but that might be a restriction imposed by the organization
343.
by woodson 9y ago
They do, but privacy is a major concern. The bigger problem is, of course, that you need speech data with (fairly) accurate transcripts for training ASR systems. These typically don't exist for call center calls.
344.
by woodson 9y ago
No, this is for speech recognition only.
345.
by woodson 9y ago
Kaldi has 4.14% WER (95.86% accuracy) on the same test dataset (test-clean) [1] using a model that runs faster than real time on CPU. You would have to make the model smaller to run it in real time on a RasPi3, but according to this [2], yo
346.
by woodson 9y ago
MATLAB actually had a lot of changes over they years that broke working code (e.g. changing the order of function arguments and outputs). It is by no means more stable than Python (2 or 3).
347.
by woodson 9y ago
Thanks! It's likely obvious to users with a technical background, but others might not realize that the STT APIs used by this service could retain uploaded audio data.
348.
by woodson 9y ago
Do you have a privacy policy? Are you storing the audio and/or edited transcripts?
349.
by woodson 9y ago
Also, check out Graham Neubig's NN for NLP class at CMU: http://phontron.com/class/nn4nlp2017/schedule.html It covers a lot of current stuff. The class is still running, so not all videos/materials are t
350.
by woodson 9y ago
Fully agree. I think their work on training data augmentation (e.g., their ICASSP paper, http://danielpovey.com/files/2017_icassp_reverberation.pdf , or the ASPiRE model before) has a bigger impact on the practical usef
351.
by woodson 9y ago
There are the CHiME challenges ( http://spandh.dcs.shef.ac.uk/chime_challenge/ ) that have test data with distant microphone speech and noisy speech data, and IARPA ran its Automatic Speech recognition In Reverberant Env
352.
by woodson 9y ago
see https://groups.google.com/d/msg/kaldi-developers/jakcCUK5lKY... It seems to be a heavily compressed version of the ASpIRE challenge example model ( https://github.com/kaldi-asr/kaldi&#
353.
by woodson 9y ago
One reason for the discrepancy between quoted numbers is that, if you are only after pushing that number down and not particularly interested in getting a scalable system, then you are free to run as many systems as you like in as many conf
354.
by woodson 9y ago
Separating acoustic model and decoding graph search makes sense since you would need a huge amount of (correctly!) transcribed speech for training. See, for example, this paper by Google [1], where they used 125,000 hours (after filtering o
355.
by woodson 9y ago
That particular implementation isn’t (FAIR research code, non-commercial), but are there known patents on the algorithms that would prevent a clean-room implementation from being used in a commercial setting?
356.
by woodson 9y ago
The situation is more difficult if you consider the possibility of having multiple contributors to a sample (mixtures), allele dropout, etc. I’ve heard some models/implementations take a day for MCMC sampling to calculate the probabili
357.
by woodson 9y ago
Snowboy isn’t open source, though.
358.
by woodson 9y ago
The technology was initially developed under the DARPA CALO research program ( https://en.m.wikipedia.org/wiki/CALO ).
359.
by woodson 9y ago
The data set is released as CC-BY-NC and, thus, cannot be used in commercial applications.
360.
by woodson 9y ago
It also works reasonably well for academic papers as PDF. Equations get butchered, of course, but overall it's not bad (you can set PDF margins so that headers/footers are not read out aloud on every page). I use it to proof-read

More ›