Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,57 @@
|
|
1 |
-
---
|
2 |
-
license: apache-2.0
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
language:
|
4 |
+
- am
|
5 |
+
- sw
|
6 |
+
- wo
|
7 |
+
tags:
|
8 |
+
- speech pretraining
|
9 |
+
- african
|
10 |
+
- fongbe
|
11 |
+
- swahili
|
12 |
+
- wolof
|
13 |
+
- wav2vec2
|
14 |
+
- ahmaric
|
15 |
+
---
|
16 |
+
|
17 |
+
# Wav2vec2-large
|
18 |
+
|
19 |
+
## Model description
|
20 |
+
This model is a pre-trained instance of the Wav2vec 2.0 architecture, specifically focused on processing and understanding four major African languages: Fongbe, Swahili,
|
21 |
+
Amharic, and Wolof. The model leverages unlabelled audio data in these languages to learn rich, language-specific representations before any fine-tuning on downstream tasks.
|
22 |
+
|
23 |
+
## Training data
|
24 |
+
|
25 |
+
The model was pre-trained using a diverse set of audio recordings from [ALFFA](https://github.com/getalp/ALFFA_PUBLIC) dataset.
|
26 |
+
|
27 |
+
* Fongbe: A Gbe language, primarily spoken in Benin and parts of Nigeria and Togo.
|
28 |
+
* Swahili: A Bantu language, widely spoken across East Africa including Tanzania, Kenya, Uganda, Rwanda, and Burundi.
|
29 |
+
* Amharic: The official language of Ethiopia, belonging to the Semitic branch of the Afroasiatic language family.
|
30 |
+
* Wolof: Predominantly spoken in Senegal, The Gambia, and Mauritania.
|
31 |
+
|
32 |
+
## Model Architecture
|
33 |
+
This model uses the large version of wav2vec 2.0 architecture developed by Facebook AI, which includes a multi-layer convolutional neural network that processes
|
34 |
+
raw audio signals to produce contextual representations. These representations are then used to predict the original audio input before any labels are provided,
|
35 |
+
following a self-supervised training methodology.
|
36 |
+
|
37 |
+
|
38 |
+
## Usage
|
39 |
+
This model is intended for use in Automatic Speech Recognition (ASR), audio classification, and other audio-related tasks in Fongbe, Swahili, Amharic, and Wolof.
|
40 |
+
To use this model for fine-tuning on a specific task, you can load it via the Hugging Face Transformers library:
|
41 |
+
|
42 |
+
```
|
43 |
+
from transformers import Wav2Vec2Processor, Wav2Vec2Model
|
44 |
+
|
45 |
+
processor = Wav2Vec2Processor.from_pretrained("your-username/wav2vec2-african-languages")
|
46 |
+
model = Wav2Vec2Model.from_pretrained("your-username/wav2vec2-african-languages")
|
47 |
+
|
48 |
+
```
|
49 |
+
|
50 |
+
## Performance
|
51 |
+
The model's performance was evaluated using a held-out validation set of audio recordings.
|
52 |
+
The effectiveness of the pre-trained representations was measured in terms of their ability to be fine-tuned to specific tasks such as ASR.
|
53 |
+
Note that detailed performance metrics will depend on the specifics of the fine-tuning process and the quality of the labeled data used.
|
54 |
+
|
55 |
+
## Limitations
|
56 |
+
The model might exhibit variability in performance across different languages due to varying amounts of training data available for each language.
|
57 |
+
Performance may degrade with audio inputs that significantly differ from the types of recordings seen during training (e.g., telephone quality audio, noisy environments).
|