[Simon]

I want an interactive way to experiment with autocorrelation in Praat

Sorry – I’m not going to implement that. But you can experiment with autocorrelation-based F0 estimation in two other ways already:

1. By playing with the settings of the pitch tracker in the unit selection exercise
2. By using the interactive example (implemented as a simple spreadsheet) in this blog post

[Simon]

Festival is not state-of-the art. Why isn’t the assignment based on Deep Neural Networks?

First of all, waveform concatenation remains dominant in commercial products (at the time of writing this post). It’s important to understand how such systems are built. Building such a system yourself is very educational in terms of script design, coverage, and so on. These issues are more opaque in DNN systems.

Second, you wouldn’t learn that much by just putting some data through a DNN training recipe.

Third, it is easy to build a DNN voice (as a side project) after building your unit selection voice.

IMPORTANT: do not build the DNN voice as part of your assignment – it is out of scope and you will NOT receive additional marks for doing this.

[Simon]

Assessment of UG and PG versions of the course

There is an UG exam because that is a 20 credit course, and a single item of assessment is not appropriate.

There is no PG exam, because we try to provide a mixed range of assessment types across different courses. A key benefit to reducing the number of exams for PG students is to ease the transition into the dissertation project component of your programme.

[Simon]

Navigating the forums is hard

There’s a lot of stuff in these forums, certainly. Use the search function – it works pretty well for queries involving one or two keywords. Browsing is also useful: you will stumble across useful things.

Note: for technical reasons, the forum search function and the website search function are separate. Navigate to a forum page, and use the “SEARCH THE FORUMS” box.

[Simon]

The coursework deadline clashes with many other deadlines

That’s very hard to avoid with an end-of-semester deadline. If there really is a pile-up of deadlines on that date (+/- 2 days), send me a list of them.

[Simon]

The course is hard if you didn’t take Speech Processing

Yes, but you were warned about that at the outset.

[Simon]

Course content, especially regarding algorithms

A few thought this very challenging. Others thought it too easy. I’m afraid that’s the nature of having such as wonderful variety of students.

If you said “too easy” then do the following

1. Complete all “extra” readings
2. Ask me for further, technical material, via the forums
3. Submit coursework that gets a mark of 100% !!

[Simon]

Lab tutoring

The presence of the lecturer in the labs was thought better than a tutor.

A couple of people thought there should be more people available in the lab. I would point out that I am idle approximately 50% of the time in lab sessions. So: please first make full use of my time before asking for this.

[Simon]

Coursework milestones

Everyone who mentioned them finds them generally helpful.

A few people wanted more detailed (and presumably therefore more frequent) milestones. But a few others thought there were too many already.

Some commented that the milestone regarding completion of the text-selection algorithm was set too early. I agree, and have tweaked that now, and as a consequence also moved the completion of all recordings a little later.

[Simon]

PDF slides do not match the videos

There was a glitch in module 6, where I posted only part of the slide pack. That is now fixed.

If you think there are other places where the slides are incomplete, please tell me.

Note that I remove redundant slides (e.g., some of the eye-candy images) to make the number of slides smaller, for those who wish to print them.

Also note that some video content is not based on slides, but is created within video-editing software (therefore, it only exists in the videos).

Please provide PDF slides on Learn

This is pointless – the slides are on speech.zone.

[Simon]

Mid-way break needed in class

The decision not to take a break in the 2-hour Speech Synthesis class was based on feedback in previous years.

I was aware that some of you have a tough timetable on Tuesdays, but hadn’t realised just how non-stop it was this year.

We will start taking a mid-way break. You will all remind me of this, if I forget!

I cannot control Informatics timetabling choices, but they have promised to look at this problem. Specifically, I’ve asked that lunchtime is left free. This will only affect future years.

[Simon]

In-class “dissection” of selected essential readings

This is very popular (much more than I expected) and so we will definitely do more of this.

[Simon]

In-class group /pair activities

The majority of you really like these and find them useful. But, a few people do not find them useful. The ratio of like vs dislike is about 2:1.

So, we’ll keep doing this, but not to excess.

[Simon]

Website is much improved

Thanks – it does work much better now. I will restructure Speech Processing for future years, along the same lines as Speech Synthesis.

[Simon]

Yes, that is correct.The particular time alignment (within each pitch period) between the detected epochs and the speech waveform will vary between algorithms. In the video, my very simple algorithm does not include any special steps to get a “good” alignment.

[Author]

Posts