41727469634d6f6e6b65793a31392e30332e32343a31373230

audio2text/a2t.py

@@ -1,8 +1,46 @@
 import librosa
 import torch
-from transformers import WhisperProcessor, WhisperForConditionalGeneration
+from __init__ import processor, model
+
+LIMIT = 90 # limit 90 seconds
 
 class A2T:
-    def __init__(self, ):
-        pass
+    def __init__(self, mic):
+        self.mic = mic
+
+    def __preproccess(self, audio, frame_rate):
+        try:
+            audio = audio / 32678.0
+
+            if len(audio.shape) > 1: 
+                audio = librosa.to_mono(audio.T)
+    
+            if frame_rate != 16_000:
+                audio = librosa.resample(audio, orig_sr=frame_rate, target_sr=16000)
+    
+            audio = audio[:16_000*LIMIT]
+    
+            audio = torch.tensor(audio)
+            return audio
+        except Exception as e:
+            print("Error", e)
+            return None
+        
+    def predict(self):
+        if this.mic is not None:
+            audio = self.mic
+            frame_rate = audio.frame_rate
+        else:
+            return "please provide audio"
+    
+        try:
+            forced_decoder_ids = processor.get_decoder_prompt_ids(language="english", task="transcribe")
+            audio = self.__preproccess(audio=audio, frame_rate=frame_rate)
+            inputs = processor(audio=audio, sampling_rate=16000, return_tensors="pt")
+            predicted_ids = model.generate(**inputs, max_length=400, forced_decoder_ids=forced_decoder_ids)
+            transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
+            return transcription[0]
+        except Exception as e:
+            print("Error", e)
+            return "Oops some kinda error"