search

znaoya / aenet

AENet: audio feature extraction
BSD 3-Clause "New" or "Revised" License
60 stars 21 forks source link

Is there a document of HCopy? #3

Closed KeyKy closed 6 years ago

KeyKy commented 7 years ago

I found that it is hard to understand the intension of following code.

cmd = self.HCopyExe + ' -C ' + self.HConfigFile + ' -S ' + scriptfile
sushanb commented 6 years ago

Bumping this thread. I am having a hard time installing the right HCopy. Any help/insights will be appreciated. Thanks Sushan

afperezm commented 6 years ago

I could only find the following reference regarding HCopy but it covers other tooling and not only HCopy. From my understanding HCopy is a tool to extract log mel-filter bank channel outputs for a given waveform audio file. In particular it uses the config file obtained by the download.sh script which contains several configurations like SOURCEKIND = WAVEFORM and TARGETKIND = FBANK_E_D_A and processes the files as specified by script file which is a temporary file containing the files to process as follows:

wav/airplane_2.wav /tmp/tmpYAHuQB/airplane_2.mfb
wav/acoustic_guitar_60.wav /tmp/tmpYAHuQB/acoustic_guitar_60.mfb

It shouldn't be necessary to download and install HCopy as it is already provided. Personally I could use as-is from Ubuntu 16.04.3 LTS.

gyglim commented 6 years ago

How to obtain HCopy, see: http://htk.eng.cam.ac.uk/download.shtml, as stated in the requirements: https://github.com/znaoya/aenet/blob/master/README.md#requirements

Indeed, we use it to extract the filter bank features, as @gantzer89 has observed.

cmd = self.HCopyExe + ' -C ' + self.HConfigFile + ' -S ' + scriptfile defines the command we use to extract the filters, with the configuration defined in self.HConfigFile, which is the following file: https://data.vision.ee.ethz.ch/cvl/aenet_feat_data/configmfb.hcopy and scriptfile being a file containing the wav files for which features are extracted