-
Notifications
You must be signed in to change notification settings - Fork 17
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Labs 2,3,4,5 submission #5
base: main
Are you sure you want to change the base?
Conversation
…nything in localhost
…NPUT" takes place
Good! Just a small comment about the time slots, it seems that the following format works: |
389c1aa
to
f89ca27
Compare
Added extra sound features (finger snap) with Audacity (tried to do it with Azure but didn't manage). Really tried my best with this one. :)
Worked on this quite awhile, had some issues with nesting some of the elements (for instance I tried to apply the <prosody pitch="low" rate="medium"></prosody> to the whole text, but for some reason it didn't work and I had to add it to each utterance separately). For some reason I also didn't hear any difference in the volume when adjusting it (for example I wanted to adjust the volume of the word "dirt" with <prosody volume="+20dB"></prosody>). Also some issues with emphasizing words, would be great if we could go through these elements once again? :)
…ach state transition!!!)
…in "Listen" state
Lab5 code works otherwise but for some reason when this guard evaluates to true: Also, I didn't quite understand this question: "Can you implement a similar threshold for natural language understanding? How can it be combined with the ASR threshold? Justify your choice (as a comment in your code) and provide a sketch implementation." since I thought we are specifically supposed to inspect the confidence levels of the NLU model's intents. |
lab 3 resource key: c6cfcc58f07142859860c92e1945fb8f |
const azureLanguageCredentials = { |
Lab 4 comments: Well done! The NLU for booking a meeting seems to be working very well. The interaction was generally smooth. The machine seems to crash if asked for a celebrity for which there is no entity on the NLU side. I worked around this by looking at the json, but it would be good to have a failsafe in place. Suggestions for improvement: Make it possible to say, for example, "Book a meeting with Vlad" directly, and get to the "at what day and time"-question. Your NLU training data suggests that this should be possible. Also, if the user only says a day in response to the "what day and time"-question, it would be good to get another question about the desired time. You can never trust your user. ;-) |
Lab 5 comments: Mostly good work, but there are two minor issues that I'd like you to fix: In case you try to assign an out-of-grammar name to a meeting, the machine replies "I couldn't properly register the meeting title, sorry." and then just stops. I'm not sure exactly why it's doing that, and it feels like uninteded behaviour. (It's not really transparent exactly what kinds of meeting titles are available. In a "real" system, it would perhaps be more flexible if the meeting titles weren't constrained by a grammar. You don't have to change this, it's more of a general comment.) If I say "Can I meet someone?" the machine just stops. There are two states named "VerifyTheTopIntentIsMeetingSpeak", which is probably the cause of this problem. It should probably be pretty easy to fix this to get a pass. |
I fixed this issue now. |
No description provided.