31–40 of 42 posts
It looks like IBM is getting serious about Watson, but still not serious enough. To create an ecosystem and incentivize developers to work through all the issues, IBM should probably create an investment fund for startups who build their products based on Watson. Any such plans?
Live demos that you can play with: Speech to Text: https://speech-to-text-demo.mybluemix.net Text to Speech: http://text-to-speech-demo.mybluemix.net
The speech synthesis is impressive. It's still clearly a computer, but the prosody is a step up from Google and Bing. I threw some random comments I've written at the English Female voice model. It seems capable of contrasting clauses via rising and falling (and handles patterns like "On the other hand, ..." and "She was either ..., or .... when ..." well), makes little dramatic pauses after noun clusters to allow th…
> it might just be good enough that I can into it on that one.
Funny. It doesn't stress as a verb, but as a noun, making the whole phrase mean something else.
Alex has a list of words like that too: live (to live) and live (live concert), progress (also verb and substantive), record, suspect and a bunch of other words that have multiple pronunciations based on the surrounding words (they are called homophones).
They should add homophone disambiguation - probably solvable with a classifier based on features extracted from surrounding words and POS tagging.
Hmm ... I along with a couple of my friends spent the last two days at the DeveloperWeek hackathon trying to explore Watson's capabilites. IBM's PaaS solution is called BlueMix and all of Watson's capabilities are available as Services for you to use. We tried using the "tradeoff Analytics" service for the project - and I must say , the tools and help available around it, the API and its documentation is pretty bad ,…
Hey..would love to talk to you and others and find out bit more around the issues..I am in sf for next few days..
1) Decouple your examples from Bluemix. It's needlessly complicating your efforts to get people using the APIs which is why people are showing up, not because they want a PAAS. At least not right off the bat. Less moving parts = better.
2) Create GOOD examples/API libraries for all the languages you're creating these examples for. The ruby on rails example is lacking. It doesn't demonstrate how to call the APIs at all. I'm using the REST API docs and the only headway I've made there is that I'm able to authenticate and create a session. Trying to upload audio using your documented approach (passing the session id as a path param) returns a 401. Passing the u/p via BasicAuth in the same request doesn't work either. As of right now I have no idea how to get an audio post to work and very little in the way of debugging information/docs.
Hmm ... I along with a couple of my friends spent the last two days at the DeveloperWeek hackathon trying to explore Watson's capabilites. IBM's PaaS solution is called BlueMix and all of Watson's capabilities are available as Services for you to use. We tried using the "tradeoff Analytics" service for the project - and I must say , the tools and help available around it, the API and its documentation is pretty bad ,…
We really appreciate the comments and will try to fix the problems. To be honest, sometimes developers can't see even the most obvious flaws in documentation. If you can highlight even one incomprehensible point it would help a lot to accelerate the revision process.
Earlier quoted context omitted.
Hi, sorry you didn't have a good experience with the API and documentation. I'd love to get some more details on what the stumbling blocks were. Thanks, jsstylos@us.ibm.com
Hmm .. OK . are there external facing portals to file bugs against the services ?
Earlier quoted context omitted.
Hey..would love to talk to you and others and find out bit more around the issues..I am in sf for next few days..
I'm in SF and spent a few hours today trying to get the speech to text API working with very little success. I'm happy to meet with you and go through this in person so you can get helpful feedback. Short of that the feedback would be to: 1) Decouple your examples from Bluemix. It's needlessly complicating your efforts to get people using the APIs which is why people are showing up, not because they want a PAAS. At l…
The ability to create your own models is very important to use this, as your existing ones do a bad job of processing my normal speech. To test I just tried reading a few simple phrases and the error rate is pretty high. The Web Speech API did a great job with the same phrases https://www.google.com/intl/en/chrome/demos/speech.html
Earlier quoted context omitted.
I'm in SF and spent a few hours today trying to get the speech to text API working with very little success. I'm happy to meet with you and go through this in person so you can get helpful feedback. Short of that the feedback would be to: 1) Decouple your examples from Bluemix. It's needlessly complicating your efforts to get people using the APIs which is why people are showing up, not because they want a PAAS. At l…
Hi..are you free on Thursday morning I can meet up at your choice of place and work through with you..
I did get things working and am working on a ruby gem that I'm hoping to be able to release for general consumption but in the mean time for folks using ruby here's a snippet that will get the simple case working once you've setup a bluemix app and bound the speech to text service to it:
There is some thing else I would like to point out as well , IBM folks on this thread can answer the question. This is regarding the "user modeling service" of Watson. I spoke to a couple of IBM folks and asked what are some of the coolest Apps they had seen that was built using Watson - and someone mentioned the following MSNBC article . It's Watson perception about the State of the Union Speech .What the user model…
We have an update coming for User Modeling (to be announced soon). After that update, such a gibberish post will return an error.
User Modeling is based on word counting. Users should ensure that their input is actually from a human and intelligible. The service looks for certain words in the input, and will reject input that doesn't have enough of those words for the service to estimate characteristics. In the upcoming release, the documentation will explain how this works and what the relevant words are.
Also, we will provide a measurement of how accurate our results are based on the number of words that are in the input. This should allow users to understand the reliability of the results in the context of their application (e.g. a casual movie recommender app might be ok with very low confidence, while an application that makes more critical recommendations might require higher confidence).