trans2

app.py

@@ -1,16 +1,17 @@
 import gradio as gr
 import torch
-from transformers import pipeline
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor, pipeline
 
 MODEL_NAME = "mikr/w2v-bert-2.0-czech-colab-cv16"
 lang = "cs"
 
 device = 0 if torch.cuda.is_available() else "cpu"
+
+model = Wav2Vec2ForCTC.from_pretrained(MODEL_NAME).to(device)
+processor = Wav2Vec2Processor.from_pretrained(MODEL_NAME)
+
 pipe = pipeline(
-    task="automatic-speech-recognition",
     model=MODEL_NAME,
-    chunk_length_s=30,
-    device=device,
 )
 
 def transcribe(file_upload):
@@ -19,14 +20,21 @@ def transcribe(file_upload):
         return "ERROR: You have to either use the microphone or upload an audio file"
 
     file = file_upload
-
     text = pipe(file)["text"]
-
     return warn_output + text
 
+def transcribe2(file_upload):
+    with torch.inference_mode():
+        input_values = processor(wav, sampling_rate=16000).input_values[0]
+        input_values = torch.tensor(input_values, device=device).unsqueeze(0)
+        logits = model(input_values).logits
+        pred_ids = torch.argmax(logits, dim=-1)
+        xcp = processor.batch_decode(pred_ids)
+        return xcp[0]
+
 
 iface = gr.Interface(
-    fn=transcribe,
+    fn=transcribe2,
     inputs=[
         gr.File(type="binary", label="Upload Audio File"),  # Audio file upload
     ],