Final changes

app.py

@@ -1,33 +1,20 @@
 import gradio as gr
 from pytube import YouTube
 from transformers import pipeline
-from transformers import WhisperProcessor, WhisperForConditionalGeneration
-import soundfile
-import os
-import subprocess
 
 class GradioInference():
   def __init__(self):
-    self.processor = WhisperProcessor.from_pretrained("humeur/whisper-small-sv-en")
-
-    self.model = WhisperForConditionalGeneration.from_pretrained("humeur/whisper-small-sv-en")
+    self.transcribe_model = pipeline(model='humeur/lab2_id2223')
+    self.translate_model = pipeline("translation_SV_to_EN", model="Helsinki-NLP/opus-mt-sv-en")
     self.yt = None
-  
+
   def __call__(self, link):
     if self.yt is None:
       self.yt = YouTube(link)
     path = self.yt.streams.filter(only_audio=True)[0].download(filename="tmp.mp4")
-    subprocess.run([
-        'ffmpeg', '-i', 'tmp.mp4','-vn', '-acodec', 'pcm_s16le', '-ac', '1', '-ar', '44100', '-f', 'wav','tmp.wav'
-    ])
-    sound_data = soundfile.read('tmp.wav')
-    input_features = self.processor(sound_data, return_tensors="pt").input_features
-    forced_decoder_ids = self.processor.get_decoder_prompt_ids(language = "sv", task = "translate")
-    predicted_ids = self.model.generate(input_features, forced_decoder_ids = forced_decoder_ids)
-    results = self.processor.batch_decode(predicted_ids, skip_special_tokens = True)
-    # results = self.model(path)
-    # return results["text"]
-    return results
+    results = self.transcribe_model(path)
+    results = self.translate_model(results["text"])
+    return results['translation_text']
 
   def populate_metadata(self, link):
     self.yt = YouTube(link)

requirements.txt

@@ -1,6 +1,5 @@
 transformers
+transformers[sentencepiece]
 pytube
 torch
 torchaudio
-sentencepiece
-soundfile