Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,11 +1,11 @@
|
|
1 |
import gradio as gr
|
2 |
-
import whisper
|
3 |
-
|
4 |
|
5 |
-
model_size = 'large-v3'
|
6 |
-
model = whisper.load_model(model_size)
|
7 |
#model = WhisperModel(model_size, device="cuda", compute_type="float16")
|
8 |
-
|
9 |
|
10 |
# or run on GPU with INT8
|
11 |
# model = WhisperModel(model_size, device="cuda", compute_type="int8_float16")
|
@@ -16,19 +16,19 @@ def speech_to_text(audio_file, _model_size):
|
|
16 |
global model_size, model
|
17 |
if model_size != _model_size:
|
18 |
model_size = _model_size
|
19 |
-
model = whisper.load_model(model_size)
|
20 |
-
|
21 |
|
22 |
-
result = model.transcribe(audio_file)
|
23 |
-
|
24 |
|
25 |
-
return result["text"]
|
26 |
-
|
27 |
|
28 |
gr.Interface(
|
29 |
fn=speech_to_text,
|
30 |
inputs=[
|
31 |
gr.Audio(source="upload", type="filepath"),
|
32 |
-
gr.Dropdown(value=model_size, choices=["tiny", "base", "small", "medium", "large", "large-v2", "large-v3"]),
|
33 |
],
|
34 |
outputs="text").launch()
|
|
|
1 |
import gradio as gr
|
2 |
+
#import whisper
|
3 |
+
from faster_whisper import WhisperModel
|
4 |
|
5 |
+
model_size = 'aka7774/whisper-large-v3-ct2'
|
6 |
+
#model = whisper.load_model(model_size)
|
7 |
#model = WhisperModel(model_size, device="cuda", compute_type="float16")
|
8 |
+
model = WhisperModel(model_size, compute_type="float16")
|
9 |
|
10 |
# or run on GPU with INT8
|
11 |
# model = WhisperModel(model_size, device="cuda", compute_type="int8_float16")
|
|
|
16 |
global model_size, model
|
17 |
if model_size != _model_size:
|
18 |
model_size = _model_size
|
19 |
+
#model = whisper.load_model(model_size)
|
20 |
+
model = WhisperModel(model_size, compute_type="float16")
|
21 |
|
22 |
+
#result = model.transcribe(audio_file)
|
23 |
+
segments, info = model.transcribe(audio_file, beam_size=5)
|
24 |
|
25 |
+
#return result["text"]
|
26 |
+
return "".join([segment.text for segment in segments])
|
27 |
|
28 |
gr.Interface(
|
29 |
fn=speech_to_text,
|
30 |
inputs=[
|
31 |
gr.Audio(source="upload", type="filepath"),
|
32 |
+
gr.Dropdown(value=model_size, choices=["tiny", "base", "small", "medium", "large", "large-v2", "large-v3", "aka7774/whisper-large-v3-ct2"]),
|
33 |
],
|
34 |
outputs="text").launch()
|