Create inference.py

inference.py

@@ -0,0 +1,31 @@
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
+import torchaudio
+import torch
+
+
+repo_id = 
+
+model = Wav2Vec2ForCTC.from_pretrained(repo_id)
+processor = Wav2Vec2Processor.from_pretrained(repo_id)
+
+
+
+def transcribe(audio_path):
+  # Load the audio file
+  audio_input, sample_rate = torchaudio.load(audio_path)
+  target_sample_rate = 16000
+  audio_input = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=target_sample_rate)(audio_input)
+
+  # Preprocess the audio data
+  input_dict = processor(audio_input[0], return_tensors="pt", padding=True, sampling_rate=16000)
+
+  # Perform inference and transcribe
+  logits = model(input_dict.input_values).logits
+  pred_ids = torch.argmax(logits, dim=-1)[0]
+  transcription = processor.decode(pred_ids)
+
+  return transcription
+
+
+# transcribe sample audio
+transcribe("download.wav")