Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,50 @@
|
|
1 |
-
---
|
2 |
-
license:
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license:
|
3 |
+
- cc-by-4.0
|
4 |
+
task_categories:
|
5 |
+
- text-to-speech
|
6 |
+
language:
|
7 |
+
- en
|
8 |
+
pretty_name: ame
|
9 |
+
size_categories:
|
10 |
+
- 10M<n<100M
|
11 |
+
configs:
|
12 |
+
- config_name: default
|
13 |
+
data_files:
|
14 |
+
- split: dev
|
15 |
+
path: data/dev-*
|
16 |
+
- split: test
|
17 |
+
path: data/test-*
|
18 |
+
- split: train
|
19 |
+
path: data/train-*
|
20 |
+
---
|
21 |
+
|
22 |
+
# Aligned English MLS
|
23 |
+
We use the Montreal Forced Aligner to align English audio files to transcripts. The dataset contains a total of 44.5K hours of English audio speech.
|
24 |
+
|
25 |
+
We only provide the aligned TextGrid files. To download the audio files, you can use the Parler MLS English dataset available at [parler-tts/mls_eng](https://huggingface.co/datasets/parler-tts/mls_eng).
|
26 |
+
|
27 |
+
This dataset can be used to train TTS models by leveraging the ground truth phoneme durations to align phoneme features with acoustic features.
|
28 |
+
|
29 |
+
### Download the Dataset
|
30 |
+
To download the dataset, use the following command:
|
31 |
+
|
32 |
+
```
|
33 |
+
huggingface-cli download ntt123/aligned_mls_eng --repo-type dataset
|
34 |
+
```
|
35 |
+
|
36 |
+
### Read TextGrid Files
|
37 |
+
To read the TextGrid files, you can use the `textgrid` library. First, install the library:
|
38 |
+
|
39 |
+
```
|
40 |
+
pip install textgrid
|
41 |
+
```
|
42 |
+
|
43 |
+
Then, load and process a TextGrid file with the following example code:
|
44 |
+
|
45 |
+
```python
|
46 |
+
import textgrid
|
47 |
+
|
48 |
+
# Load the TextGrid file
|
49 |
+
tg = textgrid.TextGrid.fromFile('test.TextGrid')
|
50 |
+
```
|