Created README
Browse files
README.md
ADDED
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
task_categories:
|
3 |
+
- audio-classification
|
4 |
+
language:
|
5 |
+
- fr
|
6 |
+
tags:
|
7 |
+
- intent
|
8 |
+
- intent-classification
|
9 |
+
- audio-classification
|
10 |
+
- audio
|
11 |
+
base_model:
|
12 |
+
- facebook/wav2vec2-xls-r-300m
|
13 |
+
datasets:
|
14 |
+
- FBK-MT/Speech-MASSIVE
|
15 |
+
library_name: transformers
|
16 |
+
license: apache-2.0
|
17 |
+
---
|
18 |
+
|
19 |
+
# wav2vec 2.0 XLS-R 128 (300m) fine-tuned on Speech-MASSIVE - fr-FR
|
20 |
+
|
21 |
+
Speech-MASSIVE is a multilingual Spoken Language Understanding (SLU) dataset comprising the speech counterpart for a portion of the MASSIVE textual corpus.
|
22 |
+
Speech-MASSIVE covers 12 languages.
|
23 |
+
It includes spoken and written utterances and is annotated with 60 intents.
|
24 |
+
The dataset is available on [HuggingFace Hub](https://huggingface.co/datasets/FBK-MT/Speech-MASSIVE).
|
25 |
+
|
26 |
+
This is the [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) model fine-tuned on the fr-FR language.
|
27 |
+
|
28 |
+
## Usage
|
29 |
+
|
30 |
+
You can use the model directly in the following manner:
|
31 |
+
|
32 |
+
```python
|
33 |
+
import torch
|
34 |
+
import librosa
|
35 |
+
from transformers import AutoModelForAudioClassification, AutoFeatureExtractor
|
36 |
+
|
37 |
+
## Load an audio file
|
38 |
+
audio_array, sr = librosa.load("path_to_audio.wav", sr=16000)
|
39 |
+
|
40 |
+
## Load model and feature extractor
|
41 |
+
model = AutoModelForAudioClassification.from_pretrained("alkiskoudounas/xls-r-128-speechmassive-fr-FR")
|
42 |
+
feature_extractor = AutoFeatureExtractor.from_pretrained("facebook/wav2vec2-xls-r-300m")
|
43 |
+
|
44 |
+
## Extract features
|
45 |
+
inputs = feature_extractor(audio_array.squeeze(), sampling_rate=feature_extractor.sampling_rate, padding=True, return_tensors="pt")
|
46 |
+
|
47 |
+
## Compute logits
|
48 |
+
logits = model(**inputs).logits
|
49 |
+
```
|