fix: fixing error to generate text

app.py

@@ -16,7 +16,7 @@ demo = gr.Interface(
                  choices=["indonesian","english"], 
                  value="indonesian"),
         gr.Audio(label="Speak", source="microphone", type="numpy"),
-        gr.Audio(label="Upload audio", source="upload", type="numpy"),
+        gr.Audio(label="Upload Audio", source="upload", type="numpy"),
     ],
     outputs=[gr.TextArea(label="Output Text"),],
     title="OpenAI Whisper Base",

src/infer.py

@@ -3,7 +3,7 @@ from typing import *
 from src import utils
 from transformers import WhisperProcessor, WhisperForConditionalGeneration
 
-model_name: str  = f"openai/whisper-base"
+model_name: str  = f"openai/whisper-small"
 processor: Any = WhisperProcessor.from_pretrained(model_name)
 model: Any = WhisperForConditionalGeneration.from_pretrained(model_name)
 

src/utils.py

@@ -7,16 +7,18 @@ sample_rate: int = 16000
 float_factor: float = 32678.0
 
 def preprocess_audio(sampling_rate, waveform):
-    waveform = waveform / float_factor
+    waveform: float = waveform / float_factor
     
-    if len(waveform) > 1:
+    if len(waveform.shape) > 1:
         waveform = librosa.to_mono(waveform.T)
     
     if sampling_rate != sample_rate:
         waveform = librosa.resample(waveform, orig_sr=sampling_rate, target_sr=sample_rate)
     
-    waveform = waveform[:sample_rate*30]
-    waveform = torch.tensor(waveform)
+    # limit to 30 seconds
+    waveform: float = waveform[:sample_rate * 30]
+    
+    waveform: float = torch.tensor(waveform)
     return waveform