Could you please offer some examples?

[Yablon]

Hello, could you please offer some exapmles, showing that how I can extract what from what ? Thank you !

[ZackHodari]

If you want to extract Merlin-equivalent labels take a look at the process_dataset.py script. It basically called lab_to_feat.py and world_with_reaper_f0.py, matches the length number of frames in the durations with the acoustic features, and trims silences (if requested).

The project is written with the intention that some people can use the provided scripts out of the box. But if you need more custom feature extraction the idea is that you can write your own script (e.g. extract phone identity).

Label creation scripts are found in lab_gen/ and have the following functionalities:

• Extract Utterance structures using Festival
• Convert Festival Utterance structures to flat HTS-style full-context labels
• Perform forced-alignment using HTK
• Extract phone-level one-hot, binary, and positional features according to a question file
• Extract frame-level counter features

Waveform feature extraction scripts can be found in wav_gen/, the script world_with_reaper_f0.py does the following:

• Extracts F0 with REAPER
• Extracts the smoothed spectrogram and band aperiodicity with WORLD
• Converts the smoothed spectrogram and band aperiodicity to Mel-scale using pySPTK

Take a look at the README for more details.

[Yablon]

@ZackHodari Thanks !