Update README.md

README.md

@@ -257,7 +257,14 @@ This version of ZeroSwot is trained with ASR data from CommonVoice, and adapting
 
 ```python
 from transformers import Wav2Vec2Processor, NllbTokenizer, AutoModel, AutoModelForSeq2SeqLM
-import soundfile as sf
+import torchaudio
+
+def load_and_resample_audio(audio_path, target_sr=16000):
+    audio, orig_freq = torchaudio.load(audio_path)
+    if orig_freq != target_sr:
+        audio = torchaudio.functional.resample(audio, orig_freq=orig_freq, new_freq=target_sr)
+    audio = audio.squeeze(0).numpy()
+    return audio
 
 # Load processors and tokenizers
 processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-large-960h-lv60-self")
@@ -277,8 +284,7 @@ nllb_model.eval()
 nllb_model.to("cuda")
 
 # Load sample .wav
-audio, sr = sf.read("sample.wav")
-assert sr == 16000, "Input of wav2vec2.0 is expected to have sampling rate of 16,000"
+audio = load_and_resample_audio("sample.wav")
 input_values = processor(audio, sampling_rate=16000, return_tensors="pt").cuda()
 
 # translation to German