Update README.md
Browse files
README.md
CHANGED
@@ -155,9 +155,9 @@ The model was trained on a **Multiaccent Catalan** speech dataset
|
|
155 |
|
156 |
### Training procedure
|
157 |
|
158 |
-
***Matxa Multiaccent Catalan***
|
159 |
|
160 |
-
The embedding layer was initialized with the number of catalan speakers per accent (
|
161 |
|
162 |
### Training Hyperparameters
|
163 |
|
|
|
155 |
|
156 |
### Training procedure
|
157 |
|
158 |
+
***Matxa Multiaccent Catalan*** is finetuned from a Catalan Matxa-base model. This Matxa-base model was finetuned from the English multispeaker checkpoint, using a 100h subset of the Catalan CommonVoice v.16 database. The selection of this small set of samples was made by using the UTMOS system, a predictor of values of the metric Mean Opinion Score (MOS) a score usually set by human evaluators according to their subjective perception of speech quality.
|
159 |
|
160 |
+
The embedding layer was initialized with the number of catalan speakers per accent (in total 8) and the original hyperparameters were kept.
|
161 |
|
162 |
### Training Hyperparameters
|
163 |
|