Update audio2text/a2t.py

audio2text/a2t.py

@@ -3,7 +3,7 @@ import numpy as np
 from .init import pipe
 
 TASK = "transcribe"
-BATCH_SIZE = 16
+BATCH_SIZE = 8
 
 class A2T:
     def __init__(self, mic):
@@ -13,32 +13,30 @@ class A2T:
         if inputs is  None:
             print("Inputs None")
 
-        transcribed_text = pipe(inputs, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)
+        transcribed_text = pipe(inputs, batch_size=BATCH_SIZE, generate_kwargs={"task": task, "language": "|en|"})
         print("transcribed_text : ", transcribed_text)
         return transcribed_text["text"]
 
     def __preprocces(self, raw: np.ndarray):
-        chunk = raw.astype(np.float32, order='C') / 32768.0
+        chunk = raw.astype(np.float32) / 32768.0
         return chunk
         
     def predict(self):
         try:
             if self.mic is not None:
                 chunk = self.mic.get_array_of_samples()
-                chunk = np.array(chunk)
+                chunk = np.array(chunk, dtype=np.int16)
                 audio = self.__preprocces(chunk)
-                sampling_rate = self.mic.frame_rate
-                print(f"audio : {audio} \n frame_rate : {sampling_rate} shape : {audio.shape}")
+                print(f"audio : {audio} \n shape : {audio.shape} \n max : {np.max(audio)}")
             else:
-                return "please provide audio"
+                raise Exception("please provide audio")
 
             if isinstance(audio , np.ndarray):
-                inputs = {"sampling_rate": 16_000, "raw": audio}
+                inputs = {"sampling_rate": 16000, "raw": audio}
                 return self.__transcribe(inputs=inputs, task=TASK)
             else:
-                return "Audio is not np array"
+                raise Exception("Audio is not np array")
                 
         except Exception as e:
-            print("Predict error", e)
-            return "Oops some kinda error"
+            return f"Oops some kinda error : {e}"