Create app.py

app.py

@@ -0,0 +1,93 @@
+import gradio as gr
+import torch
+from transformers import (
+    WhisperForConditionalGeneration,
+    WhisperTokenizer,
+    WhisperProcessor,
+    pipeline
+)
+
+# Initialize model and tokenizer
+model_id = "sudoping01/whosper-large" 
+device = "cuda" if torch.cuda.is_available() else "cpu"
+torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
+
+@gr.load(clean=True)  
+def load_model():
+    model = WhisperForConditionalGeneration.from_pretrained(
+        model_id,
+        device_map="auto",
+        use_cache=True,
+        attention_dropout=0.1,
+        dropout=0.1
+    )
+    
+    model.config.suppress_tokens = []
+    model.config.no_repeat_ngram_size = 3
+    model.config.early_stopping = True
+    model.config.max_length = 448
+    model.config.num_beams = 5
+    
+    tokenizer = WhisperTokenizer.from_pretrained(model_id)
+    processor = WhisperProcessor.from_pretrained(model_id)
+    
+    return pipeline(
+        "automatic-speech-recognition",
+        model=model,
+        tokenizer=tokenizer,
+        feature_extractor=processor.feature_extractor,
+        torch_dtype=torch_dtype,
+        chunk_length_s=30,
+        stride_length_s=3,
+        return_timestamps=False,
+        batch_size=1
+    )
+
+pipe = load_model()
+
+def transcribe(audio_path):
+    if audio_path is None:
+        return "Please provide an audio input."
+    
+    try:
+        result = pipe(
+            audio_path,
+            generate_kwargs={
+                "temperature": 0.0,
+                "do_sample": False,
+                "num_beams": 5,
+                "length_penalty": 1.0,
+                "repetition_penalty": 1.2
+            }
+        )
+        return result["text"]
+    except Exception as e:
+        return f"Error during transcription: {str(e)}"
+
+#
+css = """
+.gradio-container {max-width: 800px !important}
+.audio-btn {height: 40px; width: 40px;}
+"""
+
+demo = gr.Interface(
+    fn=transcribe,
+    inputs=[
+        gr.Audio(source="microphone", type="filepath", label="Audio Input")
+    ],
+    outputs=gr.Textbox(label="Transcription"),
+    title="Whosper Large - Speech Recognition",
+    description="Upload an audio file or record audio to transcribe speech.",
+    article="""
+    This is a demo of the Whosper Large speech recognition model.
+    - Support for multiple languages : wolof, french, english ... 
+    - Optimized for clarity and accuracy
+    - Real-time transcription capabilities
+    """,
+    examples=[], 
+    css=css,
+    enable_queue=True
+)
+
+
+demo.launch()