Spaces:
Running
on
Zero
Running
on
Zero
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import numpy as np
|
3 |
+
import torch
|
4 |
+
from peft import PeftModel, PeftConfig
|
5 |
+
from transformers import WhisperForConditionalGeneration, WhisperTokenizer, WhisperProcessor, AutomaticSpeechRecognitionPipeline
|
6 |
+
|
7 |
+
peft_model_id = "mfidabel/Modelo_1_Whisper_Large_V3"
|
8 |
+
language = "guarani"
|
9 |
+
task = "transcribe"
|
10 |
+
peft_config = PeftConfig.from_pretrained(peft_model_id)
|
11 |
+
model = WhisperForConditionalGeneration.from_pretrained(
|
12 |
+
peft_config.base_model_name_or_path, load_in_8bit=True, device_map="auto"
|
13 |
+
)
|
14 |
+
model = PeftModel.from_pretrained(model, peft_model_id)
|
15 |
+
tokenizer = WhisperTokenizer.from_pretrained(peft_config.base_model_name_or_path, language=language, task=task)
|
16 |
+
processor = WhisperProcessor.from_pretrained(peft_config.base_model_name_or_path, language=language, task=task)
|
17 |
+
feature_extractor = processor.feature_extractor
|
18 |
+
forced_decoder_ids = processor.get_decoder_prompt_ids(language="english", task=task)
|
19 |
+
|
20 |
+
pipeline = AutomaticSpeechRecognitionPipeline(model=model, tokenizer=tokenizer, feature_extractor=feature_extractor)
|
21 |
+
|
22 |
+
def transcribe(audio):
|
23 |
+
if audio is None:
|
24 |
+
return "Espera a que la grabación termine de subirse al servidor !! Intentelo de nuevo en unos segundos"
|
25 |
+
|
26 |
+
sr, y = audio
|
27 |
+
y = y.astype(np.float32)
|
28 |
+
y /= np.max(np.abs(y))
|
29 |
+
with torch.cuda.amp.autocast():
|
30 |
+
return pipeline({"sampling_rate": sr, "raw": y}, generate_kwargs={"forced_decoder_ids": forced_decoder_ids}, max_new_tokens=255)["text"]
|
31 |
+
|
32 |
+
gr.Interface(fn=transcribe, inputs="microphone", outputs="text").launch(share=True)
|