Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,68 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
---
|
| 4 |
+
`TrainingSpeech` is an initiative to provide **open and freely reusable dataset** of voices
|
| 5 |
+
|
| 6 |
+
- for speech-to-text models training
|
| 7 |
+
|
| 8 |
+
- on non-english languages
|
| 9 |
+
|
| 10 |
+
- using already available data (such as audio-books).
|
| 11 |
+
|
| 12 |
+
|
| 13 |
+
Right now, data are extracted exclusively from audio-books and in French language. Let me know if you are intersted to contribute [by creating an issue](https://github.com/wasertech/TrainingSpeech/issues/new).
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
## Tooling
|
| 18 |
+
|
| 19 |
+
`TrainingSpeech` comes with a CLI that automate and simplify:
|
| 20 |
+
- transcript extraction
|
| 21 |
+
- [forced-alignment](https://github.com/pettarin/forced-alignment-tools#definition-of-forced-alignment) (using [aeneas](https://github.com/readbeyond/aeneas))
|
| 22 |
+
- validation and correction
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
## Common workflow
|
| 27 |
+
|
| 28 |
+
### 1. Generate and validate alignment on existing source
|
| 29 |
+
|
| 30 |
+
1. pick a source that have NOT been validated yet: see `python manage.py stats` and `./sources.json` for more info
|
| 31 |
+
2. download assets (ie epub and mp3 files): `python manage.py download -s <SOURCE_NAME>`
|
| 32 |
+
3. check alignment: `python manage.py check-alignment <SOURCE_NAME>` (may require multiple iterations)
|
| 33 |
+
4. send a pull request with generated transcript and alignment
|
| 34 |
+
|
| 35 |
+
### 2. Add New source (team members only)
|
| 36 |
+
|
| 37 |
+
1. retrieve epub and corresponding mp3 file and store them into `./data/epubs` and `./data/mp3` (respectively)
|
| 38 |
+
2. create new source into `./sources.json` (NB: all fields are mandatory)
|
| 39 |
+
3. generate initial transcript using `python manage.py build-transcript <SOURCE_NAME>`
|
| 40 |
+
4. upload epub and mp3 files on S3 `python manage.py upload -s <SOURCE_NAME>`
|
| 41 |
+
|
| 42 |
+
|
| 43 |
+
## Dev setup
|
| 44 |
+
|
| 45 |
+
```sh
|
| 46 |
+
$ sudo apt-get install -y ffmpeg espeak libespeak-dev python3-numpy python-numpy libncurses-dev libncursesw5-dev sox libsqlite3-dev
|
| 47 |
+
$ git clone [email protected]:wasertech/TrainingSpeech.git
|
| 48 |
+
$ pip3 install --user pipenv
|
| 49 |
+
$ cd TrainingSpeech
|
| 50 |
+
$ pipenv install --python=3.6.6
|
| 51 |
+
$ pipenv sync
|
| 52 |
+
$ pipenv shell
|
| 53 |
+
$ pytest
|
| 54 |
+
```
|
| 55 |
+
|
| 56 |
+
|
| 57 |
+
## Last releases & download
|
| 58 |
+
Releases are ready-to-use `zip` archives containing :
|
| 59 |
+
- short 16kHz 16bit wav audio speeches (0-15s)
|
| 60 |
+
- a single `data.csv` file with following columns:
|
| 61 |
+
- `path`: path to the audio file inside the archive
|
| 62 |
+
- `duration`: audio duration in second
|
| 63 |
+
- `text`: transcript
|
| 64 |
+
|
| 65 |
+
|
| 66 |
+
| Name | # speeches | # speakers | Total Duration | Language |
|
| 67 |
+
|:--------------------------------------------------------------------------------------------------------|-------------:|-------------:|:---------------|:-----------|
|
| 68 |
+
| [2019-04-11_fr_FR](https://huggingface.co/datasets/wasertech/TrainingSpeech/resolve/main/ts_2019-04-11_fr_FR.zip?download=true) (w/ 💖 from [@lissyx](https://github.com/lissyx)) | 124089 | 4 | 182:43:35 | fr_FR |
|