abdouaziiz
MODELING_23
pulaar
tshiluba_2
tshiluba
baoule
Wav2vec2 Xls R 300m Wolof Lm
Wolof is a language spoken in Senegal and neighbouring countries, this language is not too well represented, there are few resources in the field of Text en speech In this sense we aim to bring our contribution to this, it is in this sense that enters this repo. This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m ,with a language model that is fine-tuned with the largest available speech dataset of the ALFFAPUBLIC It achieves the following results on the evaluation set: - Loss: 0.367826 - Wer: 0.212565 Model description The duration of the training data is 16.8 hours, which we have divided into 10,000 audio files for the training and 3,339 for the test. Training and evaluation data We eval the model at every 1500 step , and log it . and save at every 33340 step Training hyperparameters The following hyperparameters were used during training: - learningrate: 1e-4 - trainbatchsize: 3 - evalbatchsize : 8 - totaltrainbatchsize: 64 - totalevalbatchsize: 64 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 - lrschedulertype: linear - lrschedulerwarmupsteps: 1000 - numepochs: 10.0 | Step | Training Loss | Validation Loss | Wer | |:-------:|:-------------:|:---------------:|:------:| | 1500 | 2.854200 |0.642243 |0.543964 | | 3000 | 0.599200 | 0.468138 | 0.429549| | 4500 | 0.468300 | 0.433436 | 0.405644| | 6000 | 0.427000 | 0.384873 | 0.344150| | 7500 | 0.377000 | 0.374003 | 0.323892| | 9000 | 0.337000 | 0.363674 | 0.306189| | 10500 | 0.302400 | 0.349884 |0 .283908 | | 12000 | 0.264100 | 0.344104 |0.277120| | 13500 |0 .254000 |0.341820 |0.271316| | 15000 | 0.208400| 0.326502 | 0.260695| | 16500 | 0.203500| 0.326209 | 0.250313| | 18000 |0.159800 |0.323539 | 0.239851| | 19500 | 0.158200 | 0.310694 | 0.230028| | 21000 | 0.132800 | 0.338318 | 0.229283| | 22500 | 0.112800 | 0.336765 | 0.224145| | 24000 | 0.103600 | 0.350208 | 0.227073 | | 25500 | 0.091400 | 0.353609 | 0.221589 | | 27000 | 0.084400 | 0.367826 | 0.212565 | - Wav2vec2 + language model . - Build a Spellcheker from the text of the data - Sentence Edit Distance