Create A2T.py

audio_processing/A2T.py

@@ -0,0 +1,43 @@
+import numpy as np
+import librosa
+import io
+
+from .init import pipe
+
+TASK = "transcribe"
+BATCH_SIZE = 8
+
+class A2T:
+    def __init__(self, mic):
+        self.mic = mic
+
+    def __generate_text(self, inputs, task: str = None):
+        if inputs is  None:
+            raise Exception("Inputs is None")
+
+        transcribed_text = pipe(inputs, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)["text"]
+        return transcribed_text
+
+    def __preprocces(self, raw: bytes):
+        print(f"Raw type : {type(raw)}")
+        chunk = io.BytesIO(raw)
+        audio, sample_rate = librosa.load(chunk, sr=16000)
+        print(f"Sample rate : {sample_rate}")
+        return audio
+ 
+    def predict(self):
+        try:
+            if self.mic is not None:
+                raw = self.mic
+                audio = self.__preprocces(raw=raw)
+                print(f"audio type : {type(audio)} \n shape : {audio.shape} \n audio max value : {np.max(audio)}")
+            else:
+                raise Exception("please provide audio")
+
+            if isinstance(audio , np.ndarray):
+                return self.__generate_text(inputs=audio, task=TASK)
+            else:
+                raise Exception("Audio is not np array")
+                
+        except Exception as e:
+            return f"Oops some kinda error : {e}"