[Simon]

The approach to G2P in the video predicts exactly one phone for every letter (not a sequence of letters to a sequence of phones, as you suggest). This is a small simplification for the purposes of explanation – the video is about decision trees as a model, not G2P as a problem. A real G2P system using a CART, such as the one in Festival, would typically still process the input one letter at a time but predict 0, 1 or 2 phones. A state-of-the-art G2P model would indeed map the complete letter sequence to the complete phoneme sequence.

The video uses real data and so these frequencies are correct, as per the dictionary being used.

You suggest using data comprising word tokens from a corpus rather than word types from a dictionary. This seems reasonable, except that the goal of the CART is not to predict pronunciation for high-frequency words (they are very likely to be in the dictionary, after all), but to predict for previously unseen words.

That gives rise to the rather difficult question: what test set should we measure accuracy on? We have little choice but to hold words out of the dictionary for this purpose, but will they be representative of the yet-to-be-seen words we have to deal with later? How could we know?

So, it’s not so obvious whether types or tokens (= frequency-weighted types) make the best training data. Try it for yourself, both ways?

[Simon]

I have now added the slides to the Topic Map page.

[Simon]

The problem is almost certainly that you are not running the correct version of Festival. Make sure you have followed the instructions on changing your PATH. Post here what you see in the terminal when you type which festival.

(Try to copy-paste text rather than use a screenshot – this makes messages smaller and also searchable).

[Simon]

They are entirely separate things.

[Simon]

When typing passwords into the shell,no characters will appear (for security). But it’s still listening and will accept what you type. Just type the atlab password and hit ‘Enter’.

[Simon]

You have a lot of typos in your commands – you’re writing ‘scpl’ instead of ‘scp1’ and you are spelling rsync incorrectly.

You should copy-paste the exact command from the instructions to avoid these mistakes.

[Simon]

Glad it’s sorted. In general, we recommend not to store any valuable files inside the VM, but rather on your host computer.

[Simon]

It’s probably best not to do any software updates. If you broke it, then you could delete the whole virtual machine and download a fresh copy.

[Simon]

See this topic

[Simon]

You don’t need to export PATH if you set it in your .bashrc, just use the command in Module 3 Tutorial B.

Perhaps you have added spaces around the “=” sign?

[Simon]

This means that you have not performed synthesis for that Utterance object. Perhaps you only set the text but didn’t complete the rest of the pipeline? A Wave relation only exists in an Utterance object if the waveform generation step has been run (e.g., SayText runs all steps including that one)

[Simon]

No, you have to find mistakes: instances where the output speech is incorrect. Festival’s POS tagger is really old and not particularly good, so you will find mistakes. You might need to craft sentences in order to do this. Think about ambiguity!

[Simon]

It looks like students who enrolled slightly later on the course didn’t have their access enabled. Computing support are fixing this now…

Update: now fixed. Everyone should be able to rsync.

[Simon]

This issue with sound playback occurs in some VM installations, and can usually be solved like this.

[Simon]

You don’t need to use sudo to edit that file – it’s in your file space, so you can edit it as yourself.

You should add the PATH line at the end of the file, as per the instructions, not the start.

Don’t type the “$” before commands – this is used in the instructions to indicate the bash prompt (to differentiate from the Festival prompt).

[Author]

Posts