https://github.com/orbxball/timit-preprocessor
Extract mfcc vectors and phones from TIMIT dataset
https://github.com/orbxball/timit-preprocessor
data-preprocessing deep-learning mfcc phone speech-recognition timit timit-dataset
Last synced: 2 months ago
JSON representation
Extract mfcc vectors and phones from TIMIT dataset
- Host: GitHub
- URL: https://github.com/orbxball/timit-preprocessor
- Owner: orbxball
- License: bsd-3-clause
- Created: 2017-12-03T08:24:31.000Z (over 7 years ago)
- Default Branch: master
- Last Pushed: 2023-03-23T04:28:54.000Z (about 2 years ago)
- Last Synced: 2025-02-28T09:11:40.765Z (3 months ago)
- Topics: data-preprocessing, deep-learning, mfcc, phone, speech-recognition, timit, timit-dataset
- Language: Shell
- Size: 6.84 KB
- Stars: 15
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# TIMIT Preprocessor
**timit-preprocessor** extract mfcc vectors and phones from TIMIT dataset for advanced use on speech recognition.
## Overview
The TIMIT corpus of read speech is designed to provide speech data for acoustic-phonetic studies and for the development and evaluation of automatic speech recognition systems. More information on [website](https://catalog.ldc.upenn.edu/ldc93s1) or [Wiki](https://en.wikipedia.org/wiki/TIMIT)
## Installation
Note that to install [Kaldi](https://github.com/kaldi-asr/kaldi) first by following the instructions in [`INSTALL`](https://github.com/kaldi-asr/kaldi/blob/master/INSTALL).
> (1)
> go to tools/ and follow INSTALL instructions there.
>
> (2)
> go to src/ and follow INSTALL instructions there.
After running the scripts instructed by `INSTALL` in `tools/`, there will be reminder as followed. Go and run it.
> Kaldi Warning: IRSTLM is not installed by default anymore. If you need IRSTLM, use the script `extras/install_irstlm.sh`
After ensuring kaldi installation, we can start by running
```
git clone https://github.com/orbxball/timit-preprocessor.git
```
## Preprocessing
### Steps
1. Run `./convert_wav.sh` only in the **first time** after cloning this repo.
2. `python3 parsing.py -h` to see instructions parsing timit dataset for phone labels and raw intermediate files in folder `data/material/`.
3. `./extract_mfcc.sh` to extract mfcc vectors into .scp and .ark files.
Finally, there's a folder called `data/` which contains all the outcomes in the belowing directory structure:
```
data/
|-- material
| |-- test.lbl
| `-- train.lbl
`-- processed
|-- test.39.cmvn.ark
|-- test.39.cmvn.scp
|-- test.extract.log
|-- train.39.cmvn.ark
|-- train.39.cmvn.scp
`-- train.extract.log
```
If you want to do further operations, there's a good repo called [kaldi-io-for-python](https://github.com/vesis84/kaldi-io-for-python).
## Contact
Feel free to [contact me](mailto:[email protected]) if there's any problems.
### License
BSD 3-Clause License (2017), Jun-You Liu