Update app.py

app.py

@@ -1,11 +1,18 @@
 import gradio as gr
+import soundfile as sf
+import torch, torchaudio
 from datasets import load_dataset
+import matplotlib.pyplot as plt
 
 
 MODEL_NAME="carlosdanielhernandezmena/wav2vec2-large-xlsr-53-icelandic-ep10-1000h"
 model = Wav2Vec2ForCTC.from_pretrained(MODEL_NAME).to(device)
 processor = Wav2Vec2Processor.from_pretrained(MODEL_NAME) # do i need this? can't remember
 
+torch.random.manual_seed(0)
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+
 #def greet(name):
 #    return "Hello " + name + "!!"
 #iface = gr.Interface(fn=greet, inputs="text", outputs="text")
@@ -15,17 +22,28 @@ processor = Wav2Vec2Processor.from_pretrained(MODEL_NAME) # do i need this? can'
 
 #ds = load_dataset("language-and-voice-lab/samromur_asr",split='train',streaming=True)
 #ds = load_dataset("language-and-voice-lab/samromur_asr",split='test')
-
+#ds = ds.cast_column("audio", Audio(sampling_rate=16_000))
 
 def show_ex(exnum):
     #return(ds['audio_id'][exnum])
     return(exnum)
+    
 
-def recc(ul):
+def recc(a_f):
+    wav, sr = sf.read(a_f, dtype=np.float32)
+    if len(wav.shape) == 2:
+        wav = wav.mean(1)
+    if sr != 16000:
+        wlen = int(wav.shape[0] / sr * 16000)
+        wav = signal.resample(wav, wlen)
+        
+    with torch.inference_mode():
+        wav = torch.from_numpy(wav).unsqueeze(0)
+        if torch.cuda.is_available():
+            wav = wav.cuda()
+        input_values = processor(wav).input_values
+        return input_values
     
-    return(ul,api(ul))
-    #wait_for_model set true??
-    #anyway in a minute it timed out....
 
 bl = gr.Blocks()
 with bl:
@@ -35,6 +53,7 @@ with bl:
     #text_button.click(show_ex, inputs=text_input, outputs=text_output)
 
     audio_file = gr.Audio(type="filepath")
+    #ipt = 
     text_button.click(recc, inputs=audio_file, outputs=text_output)