Update app.py

app.py

@@ -10,7 +10,7 @@ from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Proce
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
 # load speech translation checkpoint
-asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-medium", device=device)
+asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
 
 # load text-to-speech checkpoint and speaker embeddings
 processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
@@ -21,11 +21,13 @@ vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(devic
 embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
 speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
 
+# load text en-ru translation model
+translator = pipeline("translation", model="Helsinki-NLP/opus-mt-en-ru", device=device)
 
 def translate(audio):
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "Russian"})
-    return outputs["text"]
-
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe"})
+    translated_text = translator(outputs["text"])
+    return translated_text[0]['translation_text']
 
 def synthesise(text):
     inputs = processor(text=text, return_tensors="pt")