Update app.py

app.py

@@ -4,51 +4,22 @@ from transformers import WhisperProcessor, WhisperForConditionalGeneration
 import soundfile as sf
 import numpy as np
 from scipy import signal
-import os
 
-# Set the cache directory to a writable location
-cache_dir = "/tmp/.cache"
-os.environ["TRANSFORMERS_CACHE"] = cache_dir
-os.environ["HF_DATASETS_CACHE"] = cache_dir
-os.environ["TORCH_HOME"] = cache_dir  # Set PyTorch cache directory
-
-# Ensure the cache directory exists and is writable
-os.makedirs(cache_dir, exist_ok=True)
-
-# Load the base Whisper model and processor
-def load_model():
-    print("Loading base Whisper model and processor...")
-    processor = WhisperProcessor.from_pretrained("openai/whisper-large-v3")
-    model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large-v3")
-    
-    # Load the fine-tuned weights from the Macedonian-ASR repository
-    print("Loading fine-tuned weights from Macedonian-ASR...")
-    model.load_state_dict(torch.hub.load_state_dict_from_url(
-        "https://huggingface.co/Macedonian-ASR/whisper-large-v3-macedonian-asr/resolve/main/pytorch_model.bin",
-        map_location="cpu",
-        model_dir=cache_dir  # Save downloaded weights to the writable cache directory
-    ))
-    print("✓ Model and processor loaded successfully!")
-    return processor, model
-
-processor, model = load_model()
+processor = WhisperProcessor.from_pretrained("openai/whisper-large-v3")
+model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large-v3")
 
 def process_audio(audio_path):
-    # Load and resample to 16kHz using scipy
     waveform, sr = sf.read(audio_path)
-    if len(waveform.shape) > 1:  # Convert stereo to mono
+    if len(waveform.shape) > 1:
         waveform = waveform.mean(axis=1)
-    if sr != 16000:  # Resample if necessary
+    if sr != 16000:
         num_samples = int(len(waveform) * 16000 / sr)
         waveform = signal.resample(waveform, num_samples)
     
-    # Process the audio
     inputs = processor(waveform, sampling_rate=16000, return_tensors="pt")
     predicted_ids = model.generate(**inputs, language="mk")
-    transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
-    return transcription
+    return processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
 
-# Gradio interface
 demo = gr.Interface(
     fn=process_audio,
     inputs=gr.Audio(sources=["microphone", "upload"], type="filepath"),
@@ -57,5 +28,4 @@ demo = gr.Interface(
     description="Качете аудио или користете микрофон за транскрипција на македонски говор / Upload audio or use microphone to transcribe Macedonian speech"
 )
 
-if __name__ == "__main__":
-    demo.launch()
+demo.launch(server_name="0.0.0.0", server_port=7860)