Change asr. Add translation.

app.py

@@ -9,6 +9,7 @@ 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)
+translation_pipe = pipeline("translation", model="alirezamsh/small100")
 
 # load text-to-speech checkpoint and speaker embeddings
 processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
@@ -21,9 +22,13 @@ speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze
 
 
 def translate(audio):
-    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language":"portuguese"})
+    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language":"english"})
     print('outputs', outputs)
-    return outputs["text"]
+    text = outputs["text"]
+    translation = translation_pipe(text, src_lang='pt', tgt_lang='pt')[0]
+    print('translation', translation)
+
+    return translation['translation_text']
 
 
 def synthesise(text):