Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
@@ -1,33 +1,12 @@
|
|
1 |
---
|
2 |
language:
|
3 |
- en
|
4 |
-
pretty_name: test hf dataset
|
5 |
tags:
|
6 |
- speech
|
7 |
-
license: mit
|
8 |
task_categories:
|
9 |
- text-classification
|
10 |
-
configs:
|
11 |
-
- config_name: default
|
12 |
-
data_files:
|
13 |
-
- split: test
|
14 |
-
path: data/test-*
|
15 |
-
dataset_info:
|
16 |
-
features:
|
17 |
-
- name: audio
|
18 |
-
dtype:
|
19 |
-
audio:
|
20 |
-
sampling_rate: 16000
|
21 |
-
- name: text
|
22 |
-
dtype: string
|
23 |
-
- name: time_secs
|
24 |
-
dtype: float64
|
25 |
-
splits:
|
26 |
-
- name: test
|
27 |
-
num_bytes: 230118.0
|
28 |
-
num_examples: 1
|
29 |
-
download_size: 219281
|
30 |
-
dataset_size: 230118.0
|
31 |
---
|
32 |
|
33 |
# test_hf_dataset
|
@@ -40,3 +19,42 @@ e.g.:
|
|
40 |
* contents of the dataset
|
41 |
* context for how the dataset should be used, e.g.: `datasets` package
|
42 |
* existing dataset cards, such as the ELI5 dataset card, show common conventions
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
language:
|
3 |
- en
|
4 |
+
pretty_name: "test hf dataset"
|
5 |
tags:
|
6 |
- speech
|
7 |
+
license: "mit"
|
8 |
task_categories:
|
9 |
- text-classification
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
---
|
11 |
|
12 |
# test_hf_dataset
|
|
|
19 |
* contents of the dataset
|
20 |
* context for how the dataset should be used, e.g.: `datasets` package
|
21 |
* existing dataset cards, such as the ELI5 dataset card, show common conventions
|
22 |
+
|
23 |
+
# Example usage of dataset
|
24 |
+
|
25 |
+
Example of transcription.
|
26 |
+
|
27 |
+
First install extra dependencies, typically within virtual environment.
|
28 |
+
```
|
29 |
+
python3 -m pip install datasets torch transformers
|
30 |
+
```
|
31 |
+
Then save and run this Python script. It runs transcription using the Moonshine
|
32 |
+
model by Useful Sensors [link](https://github.com/usefulsensors/moonshine).
|
33 |
+
```
|
34 |
+
"""Adapted from https://github.com/usefulsensors/moonshine#huggingface-transformers"""
|
35 |
+
from datasets import load_dataset
|
36 |
+
from transformers import AutoProcessor, MoonshineForConditionalGeneration
|
37 |
+
|
38 |
+
dataset = load_dataset("guynich/test_hf_dataset", split="test")
|
39 |
+
model = MoonshineForConditionalGeneration.from_pretrained(
|
40 |
+
"UsefulSensors/moonshine-tiny"
|
41 |
+
)
|
42 |
+
processor = AutoProcessor.from_pretrained("UsefulSensors/moonshine-tiny")
|
43 |
+
|
44 |
+
for index in range(len(dataset)):
|
45 |
+
audio_array = dataset[index]["audio"]["array"]
|
46 |
+
sampling_rate = dataset[index]["audio"]["sampling_rate"]
|
47 |
+
|
48 |
+
inputs = processor(audio_array, return_tensors="pt", sampling_rate=sampling_rate)
|
49 |
+
|
50 |
+
generated_ids = model.generate(**inputs)
|
51 |
+
|
52 |
+
transcription = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
53 |
+
print(transcription)
|
54 |
+
```
|
55 |
+
Example output.
|
56 |
+
```console
|
57 |
+
$ python3 main.py
|
58 |
+
The birch canoe slid on the smooth planks, glue the sheets to a dark blue background.
|
59 |
+
$
|
60 |
+
```
|