Hacker News Reader

Comment by devindotcom | Hacker News Reader

First video, now images - wonder if speech and others are on the way?

It's nice that they're doing this, helps advance the art I think. But it also puts a lot of smaller operations in unis sort of under the Google system in that they're best compared to Google's ML work and others using these datasets. It's a small way of stacking the deck to make Google and DeepMind more embedded in the community.

That said, its utility for others surely outweighs the strategic advantage gained here, so I for one welcome these libraries. A lot of work goes into them. Hopefully others will release theirs as well.

View on HN

I really want real-world speech data. I think speech-to-text could open the doors to lots of creative tech to help hard of hearing folks, but the dataset barrier is so massive :(

Have you seen the LibriSpeech ASR corpus[1]? It's large scale (1k hours) Creative Commons licensed English speech with the original book text.

The data is derived from read audiobooks from the LibriVox project and has been carefully segmented and aligned.

If you mean real world as in "with realistic noise", Baidu's DeepSpeech showed that adding noise to clean audio can make a resilient model. Indeed, having clean audio is better as you can add different types of noise to clean data to, in effect, create regularization.

[1]: http://www.openslr.org/12/

I wonder why semi supervised learning hasn't taken off more. There is so much unlabelled data out there. E.g. in podcasts and youtube videos and television. A small amount of it is labelled, like captioned television programs and youtube videos. You could use those captions to train a weak model which could provide labels for the unlabelled data.

You can correct many of it's errors with simple language models. For instance the phrase "wreck a nice beach" has much less probability than "recognize speech". So if the model isn't sure which one it is, you can assume it's the more probable one. Then train on that, and it will get even better at recognizing those words.

>You can correct many of it's errors with simple language models. For instance the phrase "wreck a nice beach" has much less probability than "recognize speech". So if the model isn't sure which one it is, you can assume it's the more probable one. Then train on that, and it will get even better at recognizing those words.

Why not just build that heuristic into your initial supervised "weak" model? Training on data labeled by a model introduces no new information, so you're not gaining anything there.

It does introduce new information. This is how semi-supervised learning works. Many words may be missed by the weak model, but can be inferred correctly from their context. Then you have new labels to train the weak model on, to make it better at those words.

The way I'm describing it is probably not the optimal way to do it. I don't know if there is a better way. But the point is it must be possible to take advantage of the vast quantities of unlabelled data we have. Humans brains somehow do something similar.

Semi supervised learning is really cool. I saw one successful example where they labelled just a few emails as spam and not spam. Then they trained used their weak classifier to label thousands of unclassified emails, and then used those as training data for an even stronger model. It actually works: http://matpalm.com/semi_supervised_naive_bayes/semi_supervis... https://en.wikipedia.org/wiki/Semi-supervised_learning

If you make your training set the output of Google's speech recognition algo from YT there will be lots of label noise.

It's better to go with movies and audiobooks because the text 'transcription' is human generated. Aligning known text to audio is much more accurate than speech recognition.

http://voxforge.org/

TV show audio tracks with subtitles? Not perfect, but a very, very large available dataset if you feel like downloading enough dodgy torrents. Plus points would be that any model trained would also need to be able to distinguish and separate speakers, which is something even the human brain struggles with.

Record people with a good microphone as they use their cell phone / landline (or other communication apps that compress heavily), and also record what comes out at the other end, the other end being all sorts of places, including overseas. Train a huge neural network with that, until it can extrapolate bad phone quality into something superb.

Next up: x-ray glasses, with several modes, to turn people into the hottest or ugliest possible version that could hide under those clothes.

Reply on news.ycombinator.com