arielcerdap commited on
Commit
38f6b7c
1 Parent(s): dbfdf1a

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +4 -4
app.py CHANGED
@@ -12,17 +12,17 @@ device = "cuda:0" if torch.cuda.is_available() else "cpu"
12
  asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
13
 
14
  # load text-to-speech checkpoint and speaker embeddings
15
- processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
16
 
17
- model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(device)
18
- vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
19
 
20
  embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
21
  speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
22
 
23
 
24
  def translate(audio):
25
- outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
26
  return outputs["text"]
27
 
28
 
 
12
  asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
13
 
14
  # load text-to-speech checkpoint and speaker embeddings
15
+ processor = SpeechT5Processor.from_pretrained("arielcerdap/speecht5_finetuned_voxpopuli_it")
16
 
17
+ model = SpeechT5ForTextToSpeech.from_pretrained("arielcerdap/speecht5_finetuned_voxpopuli_it").to(device)
18
+ vocoder = SpeechT5HifiGan.from_pretrained("arielcerdap/speecht5_finetuned_voxpopuli_it").to(device)
19
 
20
  embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
21
  speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
22
 
23
 
24
  def translate(audio):
25
+ outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate", "language":"it"})
26
  return outputs["text"]
27
 
28