Upload ctcalign.py

ctcalign.py

@@ -26,10 +26,10 @@ def f2s(fr):
 class CTCAligner:
     
     def __init__(self, model_path,model_word_separator, model_blank_token):
-        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        #self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
         torch.random.manual_seed(0)
         
-        self.model = Wav2Vec2ForCTC.from_pretrained(model_path).to(self.device)
+        self.model = Wav2Vec2ForCTC.from_pretrained(model_path)#.to(self.device)
         self.processor = Wav2Vec2Processor.from_pretrained(model_path)
 
         # build labels dict from a processor where it is not directly accessible
@@ -56,7 +56,7 @@ class CTCAligner:
 def get_frame_probs(wav,aligner):
     with torch.inference_mode(): # similar to with torch.no_grad():
         input_values = aligner.processor(wav,sampling_rate=16000).input_values[0]
-        input_values = torch.tensor(input_values, device=aligner.device).unsqueeze(0)
+        input_values = torch.tensor(input_values).unsqueeze(0)#, device=aligner.device).unsqueeze(0)
         emits =  aligner.model(input_values).logits
         emits = torch.log_softmax(emits, dim=-1)
     return emits[0].cpu().detach()