Update app.py

app.py

@@ -17,8 +17,7 @@ device = "cuda:0" if torch.cuda.is_available() else "cpu"
 asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device, generate_kwargs = {"task": "translate"})
 
 
-model = M2M100ForConditionalGeneration.from_pretrained("facebook/m2m100_418M")
-tokenizer = M2M100Tokenizer.from_pretrained("facebook/m2m100_418M")
+m2m100_en_sw = pipeline('translation', 'facebook/m2m100_418M', src_lang='en', tgt_lang="sw")
 
 # load text-to-speech checkpoint and speaker embeddings
 # processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
@@ -34,11 +33,12 @@ speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze
 
 def translate(audio):
     outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
-    encoded_text = tokenizer(outputs["text"], return_tensors="pt")
-    generated_tokens = model.generate(**encoded_text)
-    output_translated = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)
+    output_translated = m2m100_en_sw(outputs["text"])
+    # encoded_text = tokenizer(outputs["text"], return_tensors="pt")
+    # generated_tokens = model.generate(**encoded_text)
+    # output_translated = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)
     # return outputs["text"]
-    return output_translated
+    return output_translated["translation_text"]
 
 def synthesise(text):
     inputs = processor(text=text, return_tensors="pt")