Flexible resampling, just in case

README.md

@@ -82,7 +82,7 @@ from tqdm.auto import tqdm
 from datasets import load_metric
 from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
 
-# Download the raw data instead of using HF datasets to save space 
+# Download the raw data instead of using HF datasets to save disk space 
 data_url = "https://voice-prod-bundler-ee1969a6ce8178826482b88e843c335139bd3fb4.s3.amazonaws.com/cv-corpus-6.1-2020-12-11/sah.tar.gz"
 filestream = urllib.request.urlopen(data_url)
 data_file = tarfile.open(fileobj=filestream, mode="r|gz")
@@ -105,14 +105,13 @@ def clean_sentence(sent):
     sent = " ".join(sent.split())
     return sent
 
-resampler = torchaudio.transforms.Resample(48_000, 16_000)
-
 targets = []
 preds = []
 
 for i, row in tqdm(cv_test.iterrows(), total=cv_test.shape[0]):
     row["sentence"] = clean_sentence(row["sentence"])
     speech_array, sampling_rate = torchaudio.load(clips_path + row["path"])
+    resampler = torchaudio.transforms.Resample(sampling_rate, 16_000)
     row["speech"] = resampler(speech_array).squeeze().numpy()
 
     inputs = processor(row["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
@@ -135,4 +134,3 @@ print("WER: {:2f}".format(100 * wer.compute(predictions=preds, references=target
 
 The Common Voice `train` and `validation` datasets were used for training.
 
-The script used for training can be found [here](github.com)