[Simon]

You need to have exactly the same kind (including dimensionality) of input features at training time and at synthesis time. It sounds like you have trained a network that uses the 9 positional features, but are trying to synthesise from it without providing those features. Either

– add them to the inputs you are synthesising from

or

– remove them from the inputs for the training data

In general, the input text for synthesis should go through exactly the same processing pipeline as the text for the training data. The key difference is the use of a duration model at synthesis time, in place of forced alignment for the training data.

[Simon]

These features are apparently especially for short pauses (we think) but their origin is somewhat lost in time. I suspect you may be able to omit them with only a little degradation (try it and see for yourself though).

[Simon]

This is a matter of personal style, and there is no hard-and-fast rule. Passive is safe and formal for scientific writing, but can be boring, verbose, hard to parse and sometimes ambiguous regarding who actually came up with ideas and did the work.

First person is acceptable, although you probably don’t want to overdue this because it will read too much like a diary. So, just use first person when you need to make it clear that an idea was yours, or that you actually did the work.

A common choice in scientific writing is to use “We” instead of “I”, even for single-author papers, although this is not necessarily a good thing because it’s ambiguous about whether the author had outside help.

As always, Grammar Girl is worth a look:

http://www.quickanddirtytips.com/education/grammar/active-voice-versus-passive-voice?page=2

and her tip about using active voice without the first person (e.g. “The data suggest that…”) is excellent advice.

If you want to follow a style guide (definitely not formally required for your dissertation!), then APA is a good default

http://www.apastyle.org

[Simon]

You need to append the bottleneck features at synthesis time too. So, synthesis will also involve a forward pass through the bottleneck network, saving those bottleneck features, appending them to the usual input features, then passing this concatenated vector through the second network.

[Simon]

Looks like you have spaces in some of your directory names

DNN Acoustic models

DNN try WORLD

This is a bad idea because it will break many shell scripts or system calls. Replace those spaces with underscores.

[Simon]

yes, of course – but I can’t do this

use the Informatics support form to request more space (say that you are my student)

[Simon]

A possible explanation is filesystem / network glitches – are you working on the iMacs in your Documents folder? If so, try making a copy of everything on to the machine’s own local disk and running there.

Simon

[Simon]

This would seem to be the error:

/this/path/does/not/exist/x2x

Looks like you need to set a correct path to the SPTK tools (of which x2x is a part)

[Simon]

The fix is easy, and just needs a simple improvement to the code. One obvious fix would be to wrap a “try….except” around the os.makedirs function call. The exception that you need to catch (and then ignore, perhaps printing an informational message to the log) is OSError

[Simon]

Have you run out of disk quota (are you working on DICE or the iMacs)?

[Simon]

Do you get any error message?

[Simon]

If there is no trained model, then check that you have actually run that stage:

TRAINDNN : True

[Simon]

Remember that you can modify (and improve!) the code. Here, it looks like you should add some debugging around the place where the error occurs. For example, print out the sox command just before executing it.

[Simon]

First, you cannot upgrade the copy in /Volumes/Network/courses/ss/dnn/dnn_tts/ – that is not owned by you!

You need to check out your own copy. If you still get the error, then do

$ svn upgrade

then

$ svn update

[Simon]

There is a working version of bandmat here

/Volumes/Network/courses/ss/dnn/dnn_tts/bandmat

[Author]

Posts