Commit
•
bcec9d9
1
Parent(s):
6c9ca2c
add readme
Browse files
README.md
ADDED
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
---
|
3 |
+
license: cc-by-4.0
|
4 |
+
task_categories:
|
5 |
+
- automatic-speech-recognition
|
6 |
+
language:
|
7 |
+
- en
|
8 |
+
pretty_name: People's Speech Clean
|
9 |
+
---
|
10 |
+
|
11 |
+
# Distil Whisper: People's Speech Clean
|
12 |
+
|
13 |
+
This is a variant of the [People's Speech Clean](https://huggingface.co/datasets/MLCommons/peoples_speech) dataset, augmented to return the pseudo-labelled Whisper
|
14 |
+
Transcriptions alongside the original dataset elements. The pseudo-labelled transcriptions were generated by
|
15 |
+
labelling the input audio data with the Whisper [large-v2](https://huggingface.co/openai/whisper-large-v2)
|
16 |
+
model with *greedy* sampling. For information on how the original dataset was curated, refer to the original
|
17 |
+
[dataset card](https://huggingface.co/datasets/MLCommons/peoples_speech).
|
18 |
+
|
19 |
+
## Standalone Usage
|
20 |
+
|
21 |
+
First, install the latest version of the 🤗 Datasets package:
|
22 |
+
|
23 |
+
```bash
|
24 |
+
pip install --upgrade pip
|
25 |
+
pip install --upgrade datasets[audio]
|
26 |
+
```
|
27 |
+
|
28 |
+
The dataset can be downloaded and pre-processed on disk using the [`load_dataset`](https://huggingface.co/docs/datasets/v2.14.5/en/package_reference/loading_methods#datasets.load_dataset)
|
29 |
+
function:
|
30 |
+
|
31 |
+
```python
|
32 |
+
from datasets import load_dataset
|
33 |
+
|
34 |
+
dataset = load_dataset("distil-whisper/peoples_speech-clean", "clean")
|
35 |
+
# take the first sample of the validation set
|
36 |
+
sample = dataset["validation"][0]
|
37 |
+
```
|
38 |
+
|
39 |
+
It can also be streamed directly from the Hub using Datasets' [streaming mode](https://huggingface.co/blog/audio-datasets#streaming-mode-the-silver-bullet).
|
40 |
+
Loading a dataset in streaming mode loads individual samples of the dataset at a time, rather than downloading the entire
|
41 |
+
dataset to disk:
|
42 |
+
|
43 |
+
```python
|
44 |
+
from datasets import load_dataset
|
45 |
+
|
46 |
+
dataset = load_dataset("distil-whisper/peoples_speech-clean", "clean", streaming=True)
|
47 |
+
# take the first sample of the validation set
|
48 |
+
sample = next(iter(dataset["validation"]))
|
49 |
+
```
|
50 |
+
|
51 |
+
## Distil Whisper Usage
|
52 |
+
|
53 |
+
To use this dataset to reproduce a Distil Whisper training run, refer to the instructions on the
|
54 |
+
[Distil Whisper repository](https://github.com/huggingface/distil-whisper#training).
|
55 |
+
|
56 |
+
## License
|
57 |
+
|
58 |
+
This dataset is licensed under
|
59 |
+
|