Improved results using a 8s + 2s chunking strategy

README.md

@@ -38,10 +38,10 @@ model-index:
     metrics:
        - name: Test WER
          type: wer
-         value: 22.58
+         value: 20.79
        - name: Test CER
          type: cer
-         value: 11.26
+         value: 10.72
 ---
 
 # XLS-R-based CTC model with 5-gram language model from Common Voice

eval.sh

@@ -1,2 +1,2 @@
 python ./eval.py --model_id FremyCompany/xls-r-nl-v1-cv8-lm --dataset mozilla-foundation/common_voice_8_0 --config nl --split test --log_outputs
-python ./eval.py --model_id FremyCompany/xls-r-nl-v1-cv8-lm --dataset speech-recognition-community-v2/dev_data --config nl --split validation --chunk_length_s 5.0 --stride_length_s 1.0
+python ./eval.py --model_id FremyCompany/xls-r-nl-v1-cv8-lm --dataset speech-recognition-community-v2/dev_data --config nl --split validation --chunk_length_s 8.0 --stride_length_s 2.0

speech-recognition-community-v2_dev_data_nl_validation_eval_results.txt

@@ -1,2 +1,2 @@
-WER: 0.225885510882885
-CER: 0.11266761318820002
+WER: 0.20791270350703156
+CER: 0.10724088973024136