Malaya-Speech Cache
Contents
Malaya-Speech Cache#
This tutorial is available as an IPython notebook at malaya-speech/example/caching.
This module only useful if use default Malaya-Speech repository, for huggingface, read more at https://huggingface.co/docs/datasets/v1.12.0/cache.html
Default Cache location#
You actually can know where is your Malaya default caching folder. Caching folder will save any models, vocabs, and rules downloaded for specific modules.
[1]:
import malaya_speech
[3]:
malaya_speech.__home__
[3]:
'/Users/huseinzolkepli/Malaya-Speech'
Change default Cache location#
To change default cache location, you need to set MALAYA_SPEECH_CACHE
OS environment before import Malaya-Speech,
export MALAYA_SPEECH_CACHE=/Users/huseinzolkepli/Documents/malaya-speech-cache
Or you can set in bashenv to make it permanent if you want.
[4]:
import os
os.environ['MALAYA_CACHE'] = '/Users/huseinzolkepli/Documents/malaya-speech-cache'
Cache subdirectories#
Malaya-Speech put models in subdirectories, you can print it by simply,
[5]:
malaya_speech.utils.print_cache()
Malaya-Speech/
├── gender/
│ ├── deep-speaker/
│ │ ├── model.pb
│ │ └── version
│ ├── vggvox-v2/
│ │ ├── model.pb
│ │ └── version
│ └── vggvox-v2-quantized/
│ ├── model.pb
│ └── version
├── language-model/
│ ├── bahasa/
│ │ ├── ctc-bahasa.json
│ │ ├── model.trie.klm
│ │ └── version
│ └── bahasa-combined/
│ ├── ctc-bahasa.json
│ ├── model.trie.klm
│ └── version
├── multispeaker-separation-wav/
│ ├── fastsep-4/
│ │ ├── model.pb
│ │ └── version
│ └── fastsep-4-quantized/
│ ├── model.pb
│ └── version
├── speaker-vector/
│ └── vggvox-v2/
│ ├── model.pb
│ └── version
├── speech-to-text/
│ ├── conformer-mixed/
│ │ ├── model.pb
│ │ ├── transducer-mixed.subword.subwords
│ │ └── version
│ ├── conformer-mixed-quantized/
│ │ ├── model.pb
│ │ ├── transducer-mixed.subword.subwords
│ │ └── version
│ ├── small-conformer-mixed/
│ │ ├── model.pb
│ │ ├── transducer-mixed.subword.subwords
│ │ └── version
│ └── small-conformer-mixed-quantized/
│ ├── model.pb
│ ├── transducer-mixed.subword.subwords
│ └── version
├── speech-to-text-ctc/
│ ├── wav2vec2-conformer/
│ │ ├── ctc-bahasa.json
│ │ ├── model.pb
│ │ └── version
│ └── wav2vec2-conformer-quantized/
│ ├── ctc-bahasa.json
│ ├── model.pb
│ └── version
├── super-resolution/
│ ├── srgan-128/
│ │ ├── model.pb
│ │ └── version
│ └── srgan-256/
│ ├── model.pb
│ └── version
├── tts/
│ ├── fastspeech2-female-singlish/
│ │ ├── model.pb
│ │ ├── quantized/
│ │ │ ├── model.pb
│ │ │ └── version
│ │ └── version
│ ├── stats/
│ │ └── female-singlish.npy
│ └── tacotron2-female-singlish/
│ ├── model.pb
│ ├── quantized/
│ │ ├── model.pb
│ │ └── version
│ └── version
├── vad/
│ └── vggvox-v2/
│ ├── model.pb
│ └── version
├── version
└── vocoder-melgan/
├── female-singlish/
│ ├── model.pb
│ └── version
├── universal-1024/
│ ├── model.pb
│ └── version
└── universal-1024-quantized/
├── model.pb
└── version
Deleting specific model#
you can specifically choose which model you want to delete.
[6]:
malaya_speech.utils.delete_cache('language-model/bahasa-combined')
[6]:
True
What happen if a directory does not exist?
[9]:
malaya_speech.utils.delete_cache('language-model/bahasa-combined2')
---------------------------------------------------------------------------
Exception Traceback (most recent call last)
<ipython-input-9-328c5a06876b> in <module>
----> 1 malaya_speech.utils.delete_cache('language-model/bahasa-combined2')
~/Documents/tf-1.15/env/lib/python3.7/site-packages/malaya_boilerplate-0.0.3-py3.7.egg/malaya_boilerplate/utils.py in delete_cache(location)
188 if not os.path.exists(location):
189 raise Exception(
--> 190 f'folder not exist, please check path from `{__package__}.utils.print_cache()`'
191 )
192 if not os.path.isdir(location):
Exception: folder not exist, please check path from `malaya-speech.utils.print_cache()`
Purge cache#
You can simply delete all models, totally purge it. By simply,
[10]:
malaya_speech.utils.delete_all_cache
[10]:
<function malaya_boilerplate.utils.delete_all_cache()>
I am not gonna to run it, because I prefer to keep it for now?