Update README.md

Changed evaluation script resampler to librosa which was used in the model training. This improves the WER score a little bit.

README.md

@@ -23,7 +23,7 @@ model-index:
     metrics:
        - name: Test WER
          type: wer
-         value: 32.607866
+         value: 32.378771
 ---
 
 # Wav2Vec2-Large-XLSR-53-Finnish
@@ -74,6 +74,7 @@ The model can be evaluated as follows on the Finnish test data of Common Voice.
 
 
 ```python
+import librosa
 import torch
 import torchaudio
 from datasets import load_dataset, load_metric
@@ -88,14 +89,14 @@ model = Wav2Vec2ForCTC.from_pretrained("aapot/wav2vec2-large-xlsr-53-finnish")
 model.to("cuda")
 
 chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“\%\‘\”\�\'\...\…\–\é]'
-resampler = lambda sr: torchaudio.transforms.Resample(sr, 16_000)
+resampler = lambda sr: lambda y: librosa.resample(y.numpy().squeeze(), sr, 16_000)
 
 # Preprocessing the datasets.
 # We need to read the audio files as arrays
 def speech_file_to_array_fn(batch):
   batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
   speech_array, sampling_rate = torchaudio.load(batch["path"])
-  batch["speech"] = resampler(sampling_rate)(speech_array).squeeze().numpy()
+  batch["speech"] = resampler(sampling_rate)(speech_array).squeeze()
   return batch
 
 test_dataset = test_dataset.map(speech_file_to_array_fn)
@@ -117,7 +118,7 @@ result = test_dataset.map(evaluate, batched=True, batch_size=8)
 print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
 ```
 
-**Test Result**: 32.607866 %
+**Test Result**: 32.378771 %
 
 
 ## Training