Spaces:
Runtime error
Runtime error
Commit
•
08572fb
1
Parent(s):
c22314e
Update app/main.py
Browse files- app/main.py +36 -1
app/main.py
CHANGED
@@ -1,8 +1,43 @@
|
|
1 |
from fastapi import FastAPI
|
2 |
-
|
|
|
3 |
|
4 |
app = FastAPI()
|
5 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
@app.get("/")
|
7 |
def root():
|
8 |
return {"Hello": "World"}
|
|
|
1 |
from fastapi import FastAPI
|
2 |
+
import torch
|
3 |
+
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
|
4 |
|
5 |
app = FastAPI()
|
6 |
|
7 |
+
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
8 |
+
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
9 |
+
|
10 |
+
model_id = "openai/whisper-large-v3"
|
11 |
+
|
12 |
+
model = AutoModelForSpeechSeq2Seq.from_pretrained(
|
13 |
+
model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
|
14 |
+
)
|
15 |
+
model.to(device)
|
16 |
+
|
17 |
+
processor = AutoProcessor.from_pretrained(model_id)
|
18 |
+
|
19 |
+
pipe = pipeline(
|
20 |
+
"automatic-speech-recognition",
|
21 |
+
model=model,
|
22 |
+
tokenizer=processor.tokenizer,
|
23 |
+
feature_extractor=processor.feature_extractor,
|
24 |
+
max_new_tokens=128,
|
25 |
+
chunk_length_s=30,
|
26 |
+
batch_size=16,
|
27 |
+
return_timestamps=True,
|
28 |
+
torch_dtype=torch_dtype,
|
29 |
+
device=device,
|
30 |
+
)
|
31 |
+
|
32 |
+
def speech_to_text(path_to_file: str) -> str:
|
33 |
+
"""
|
34 |
+
Given the path to the .wav file, it returns the transcription.
|
35 |
+
"""
|
36 |
+
result = pipe(path_to_file)
|
37 |
+
|
38 |
+
return result["text"]
|
39 |
+
|
40 |
+
|
41 |
@app.get("/")
|
42 |
def root():
|
43 |
return {"Hello": "World"}
|