Update README.md

README.md

@@ -11,37 +11,61 @@ probably proofread and complete it, then remove this comment. -->
 
 # wav2vec2-xls-r-300m-mixed
 
-This model was trained from scratch on an unknown dataset.
-It achieves the following results on the evaluation set:
+Finetuned https://huggingface.co/facebook/wav2vec2-xls-r-300m on https://github.com/huseinzol05/malaya-speech/tree/master/data/mixed-stt
 
+This model was finetuned on 3 languages,
 
-## Model description
+1. Malay
+2. Singlish
+3. Mandarin
 
-More information needed
+**This model trained on a single RTX 3090 Ti 24GB VRAM, provided by https://mesolitica.com/**. 
 
-## Intended uses & limitations
+## Evaluation set
 
-More information needed
+Evaluation set from https://github.com/huseinzol05/malaya-speech/tree/master/pretrained-model/prepare-stt with sizes,
 
-## Training and evaluation data
+```
+len(malay), len(singlish), len(mandarin)
+-> (765, 3579, 614)
+```
 
-More information needed
+It achieves the following results on the evaluation set based on [evaluate-wav2vec2-xls-r-300m-mixed.ipynb](evaluate-wav2vec2-xls-r-300m-mixed.ipynb):
 
-## Training procedure
+Mixed evaluation,
 
-### Training hyperparameters
+```
+CER: 0.05082346216269688
+WER: 0.14251665517797765
+CER with LM: 0.042868860764264445
+WER with LM: 0.10380217528405207
+```
 
-The following hyperparameters were used during training:
-- optimizer: None
-- training_precision: float32
+Malay evaluation,
 
-### Training results
+```
+CER: 0.05027226867066105
+WER: 0.21723938552369926
+CER with LM: 0.03601546154013878
+WER with LM: 0.13593624603428525
+```
 
+Singlish evaluation,
 
+```
+CER: 0.05161275767676772
+WER: 0.1331819722523124
+CER with LM: 0.04419848182804781
+WER with LM: 0.09859626021111582
+```
 
-### Framework versions
+Mandarin evaluation,
 
-- Transformers 4.18.0
-- TensorFlow 2.6.0
-- Datasets 2.1.0
-- Tokenizers 0.12.1
+```
+CER: 0.04690941391603209
+WER: 0.10382926344585862
+CER with LM: 0.0436573568867001
+WER with LM: 0.09411065398455744
+```
+
+Language model from https://huggingface.co/huseinzol05/language-model-bahasa-manglish-combined