correcting CER score

README.md

@@ -21,10 +21,14 @@ model-index:
     metrics:
        - name: Test CER
          type: cer
-         value: 76.57
+         value: 16.41
 ---
 
-## Colab trial with recording or voice file
+# Wav2Vec2-Large-XLSR-53-hk
+Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) on Cantonese using the [Common Voice](https://huggingface.co/datasets/common_voice).
+When using this model, make sure that your speech input is sampled at 16kHz.
+
+## Usage
 [Colab trial](https://colab.research.google.com/drive/1nBRLf4Pwiply_y5rXWoaIB8LxX41tfEI?usp=sharing)
 
 ```
@@ -73,8 +77,15 @@ Predict
 predict(load_file_to_data('voice file path'))
 ```
 
-## Evaluation on Common Voice HK Test
+## Evaluation
+The model can be evaluated as follows on the Chinese (Hong Kong) test data of Common Voice.   
+CER calculation refer to https://huggingface.co/ctl/wav2vec2-large-xlsr-cantonese 
+
 ```python
+!mkdir cer
+!wget -O cer/cer.py https://huggingface.co/ctl/wav2vec2-large-xlsr-cantonese/raw/main/cer.py
+!pip install jiwer
+
 import torchaudio
 from datasets import load_dataset, load_metric
 from transformers import (
@@ -85,6 +96,7 @@ import torch
 import re
 import sys
 
+cer = load_metric("./cer")
 model_name = "voidful/wav2vec2-large-xlsr-53-hk"
 device = "cuda"
 processor_name = "voidful/wav2vec2-large-xlsr-53-hk"
@@ -94,7 +106,7 @@ chars_to_ignore_regex = r"[¥•＂＃＄％＆＇（）＊＋，－／：；＜
 model = Wav2Vec2ForCTC.from_pretrained(model_name).to(device)
 processor = Wav2Vec2Processor.from_pretrained(processor_name)
 
-ds = load_dataset("common_voice", 'zh-HK', split="test")
+ds = load_dataset("common_voice", 'zh-HK', data_dir="./cv-corpus-6.1-2020-12-11", split="test")
 
 resampler = torchaudio.transforms.Resample(orig_freq=48_000, new_freq=16_000)
 
@@ -120,9 +132,7 @@ def map_to_pred(batch):
 
 result = ds.map(map_to_pred, batched=True, batch_size=16, remove_columns=list(ds.features.keys()))
 
-wer = load_metric("wer")
-
-print(wer.compute(predictions=result["predicted"], references=result["target"]))
+print("CER: {:2f}".format(100 * cer.compute(predictions=result["predicted"], references=result["target"])))
 ```
 
-`CER 0.7656911821332306`
+`CER 16.41`