example missing

[derNarr]

Hi,
a very short example would be nice to see and some documentation possibly coupled with readthedocs would be really nice.
Tino

p.s. I do it with sphinx, for an example look here https://github.com/quantling/pyndl/ and https://pyndl.readthedocs.io/en/latest/development.html (Nowadays there might be different / better ways to do it.)

[paul-krug]

Hi Tino,
I am in the process of preparing a detailed documentation with readthedocs. I hope I can release it within the next month.
I will try to add some examples as soon as possible.

Cheers,
Paul

[BucketofJava]

Any updates on this?

[paul-krug]

Hi, unfortunately there is still no detailed documentation because I may re-structure larger parts of the code in a future release.

Anyway, I added a pdf file with some slides that explain the basic objects and functions:
https://github.com/paul-krug/VocalTractLab-Python/tree/master/VocalTractLab/examples

I hope this helps. Let me know if anything is unclear, or if you need help with a specific problem.

Cheers,
Paul

[BucketofJava]

Thank you so much!

[BucketofJava]

One question I do have is what is a tract sequence and how does it differ from a gestural score, as I didn't find anything about tract sequences in the manual?
Thanks so much

[derNarr]

Very roughly speaking the gestural scores define targets, duration, and relaxations time on the following eight tiers:

• Vowel gesture
• Lip Gesture
• Tongue tip
• Tongue body
• Velic Gestures
• Glottal Shape
• F0 gestures
• Lung pressure

whereas the tract sequence defines the 19 vocal tract parameters and 11 glottis parameters every 110 samples of the synthesized 44100 Hz mono signal (roughly every 2.5 Milliseconds). The tract shape is linearly interpolated between these points and no duration or targets are defined.

It is possible to derive a tract sequence from gestural scores, but not the other way around. Therefore the tract sequence can be seen as the geometrical realization of the geometrical model in the VTL.

@paul-krug correct me, if I misstated anything.

@BucketofJava if you are more interested in a ctypes access to the VTL API have a look at the Python examples in https://github.com/TUD-STKS/VocalTractLabBackend-dev/tree/main/examples

[paul-krug]

Thats not entirely correct, as tract sequences have a different meaning in VTL-Python.
These objects are basically ensembles of VTL motor states. They can be directly synthesized as described above. However, they can also be turned into a Motor Score (by interpreting the single states as target states). In that case, the interpolation between these motor states happens within a given timeframe based on the Target-Approximation-Model (see e.g. https://discovery.ucl.ac.uk/id/eprint/10071664/7/Xu%20birkholz-2018-eusipco.pdf for some details on target approximation).

Gestural scores on the other hand are high-level representations of motor scores. In VTL-Python they are no objects, but will be exported or processed via file output/input. You can create them from segment sequences (sequences of phonemes) and you can turn them to tract sequence files or audio.

I do not recommend the use of ctypes access to the backend as it has numerous disadvantages, especially if you aim for multiprocessing.

Cheers,
Paul

[BucketofJava]

Thank you!