Commit
•
3bfb858
1
Parent(s):
1025309
Update app.py
Browse files
app.py
CHANGED
@@ -18,8 +18,8 @@ model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts").to(dev
|
|
18 |
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
|
19 |
|
20 |
|
21 |
-
|
22 |
-
|
23 |
|
24 |
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
|
25 |
speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
|
@@ -31,16 +31,16 @@ def translate(audio):
|
|
31 |
|
32 |
|
33 |
def synthesise(text):
|
34 |
-
inputs =
|
35 |
input_ids = inputs["input_ids"]
|
36 |
|
37 |
|
38 |
with torch.no_grad():
|
39 |
-
outputs =
|
40 |
-
print(outputs)
|
41 |
inputs = processor(text=text, return_tensors="pt")
|
42 |
speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
|
43 |
-
print(speech)
|
44 |
return speech.cpu()
|
45 |
|
46 |
|
|
|
18 |
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
|
19 |
|
20 |
|
21 |
+
model_mms = VitsModel.from_pretrained("Matthijs/mms-tts-deu").to(device)
|
22 |
+
tokenizer_mms = VitsTokenizer.from_pretrained("Matthijs/mms-tts-deu")
|
23 |
|
24 |
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
|
25 |
speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
|
|
|
31 |
|
32 |
|
33 |
def synthesise(text):
|
34 |
+
inputs = tokenizer_mms(text, return_tensors="pt")
|
35 |
input_ids = inputs["input_ids"]
|
36 |
|
37 |
|
38 |
with torch.no_grad():
|
39 |
+
outputs = model_mms(input_ids)
|
40 |
+
print("mms model", outputs)
|
41 |
inputs = processor(text=text, return_tensors="pt")
|
42 |
speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
|
43 |
+
print("speecht5 model", speech)
|
44 |
return speech.cpu()
|
45 |
|
46 |
|