Rifat Mamayusupov
commited on
Commit
•
39cc8c0
1
Parent(s):
32438e9
Update README.md
Browse files
README.md
CHANGED
@@ -20,17 +20,53 @@ It achieves the following results on the evaluation set:
|
|
20 |
|
21 |
## Model description
|
22 |
|
23 |
-
|
24 |
|
25 |
-
|
26 |
|
27 |
-
|
|
|
|
|
|
|
28 |
|
29 |
-
|
30 |
|
31 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
|
33 |
-
## Training procedure
|
34 |
|
35 |
### Training hyperparameters
|
36 |
|
|
|
20 |
|
21 |
## Model description
|
22 |
|
23 |
+
UZBTTS - bu asason 250 MB Text2Audio datasetga (microsoft/speecht5_tts) modeliga fine-tuned qilindi, natija datasetga yarasha yaxshi.
|
24 |
|
25 |
+
Agar siz buni modelni foydalanishini xoxlasangiz.
|
26 |
|
27 |
+
example:
|
28 |
+
```
|
29 |
+
#dastlab run qiling :
|
30 |
+
!pip install transformers datasets
|
31 |
|
32 |
+
from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech
|
33 |
|
34 |
+
processor = SpeechT5Processor.from_pretrained("ai-nightcoder/UZBTTS")
|
35 |
+
model = SpeechT5ForTextToSpeech.from_pretrained("ai-nightcoder/UZBTTS")
|
36 |
+
|
37 |
+
# ***************************************************************************
|
38 |
+
text = "O‘zbekistonda import qilingan sovitkich,
|
39 |
+
muzlatkich va konditsionerlarni energosamaradorlik bo‘yicha sinovdan o‘tkazish boshlandi.
|
40 |
+
Kun.uz'ga murojaat qilgan importchi tadbirkorlarga ko‘ra, bu yangilik ham vaqt,
|
41 |
+
ham naqd nuqtayi nazaridan yangi xarajatlarga olib kelgan.
|
42 |
+
Kelgusida bunday tekshiruv boshqa turdagi maishiy texnikalarga ham joriy etilishi kutilyapti."
|
43 |
+
|
44 |
+
inputs = processor(text=text, return_tensors="pt")
|
45 |
+
|
46 |
+
# ***************************************************************************
|
47 |
+
|
48 |
+
from datasets import load_dataset
|
49 |
+
|
50 |
+
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
|
51 |
+
|
52 |
+
import torch
|
53 |
+
# voice clone uchun ham ishlatilsa bo'ladi.
|
54 |
+
speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
|
55 |
+
|
56 |
+
spectrogram = model.generate_speech(inputs["input_ids"], speaker_embeddings)
|
57 |
+
|
58 |
+
from transformers import SpeechT5HifiGan
|
59 |
+
|
60 |
+
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")
|
61 |
+
|
62 |
+
# ****************************************************************************
|
63 |
+
|
64 |
+
speech = model.generate_speech(inputs["input_ids"], speaker_embeddings, vocoder=vocoder)
|
65 |
+
|
66 |
+
from IPython.display import Audio
|
67 |
+
|
68 |
+
Audio(speech, rate=16000)
|
69 |
|
|
|
70 |
|
71 |
### Training hyperparameters
|
72 |
|