Spaces:
Running
on
Zero
Running
on
Zero
adrian-saez-martinez
commited on
Commit
β’
f11f43f
1
Parent(s):
e2c56f2
app.py added
Browse files
app.py
ADDED
@@ -0,0 +1,93 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import gradio as gr
|
3 |
+
from transformers import pipeline
|
4 |
+
import concurrent.futures
|
5 |
+
import time
|
6 |
+
|
7 |
+
# Load both models
|
8 |
+
MODEL_NAME_TURBO = "openai/whisper-large-v3-turbo"
|
9 |
+
MODEL_NAME_STANDARD = "openai/whisper-large-v3"
|
10 |
+
|
11 |
+
device = 0 if torch.cuda.is_available() else "cpu"
|
12 |
+
|
13 |
+
# Set up the pipeline for both models
|
14 |
+
pipe_turbo = pipeline(
|
15 |
+
task="automatic-speech-recognition",
|
16 |
+
model=MODEL_NAME_TURBO,
|
17 |
+
chunk_length_s=30,
|
18 |
+
device=device,
|
19 |
+
)
|
20 |
+
|
21 |
+
pipe_standard = pipeline(
|
22 |
+
task="automatic-speech-recognition",
|
23 |
+
model=MODEL_NAME_STANDARD,
|
24 |
+
chunk_length_s=30,
|
25 |
+
device=device,
|
26 |
+
)
|
27 |
+
|
28 |
+
# Function to transcribe audio using the turbo model
|
29 |
+
def transcribe_turbo(audio):
|
30 |
+
start_time = time.time()
|
31 |
+
text_turbo = pipe_turbo(audio)["text"]
|
32 |
+
elapsed_time = time.time() - start_time
|
33 |
+
return text_turbo, elapsed_time
|
34 |
+
|
35 |
+
# Function to transcribe audio using the standard model
|
36 |
+
def transcribe_standard(audio):
|
37 |
+
start_time = time.time()
|
38 |
+
text_standard = pipe_standard(audio)["text"]
|
39 |
+
elapsed_time = time.time() - start_time
|
40 |
+
return text_standard, elapsed_time
|
41 |
+
|
42 |
+
# Function to compare transcriptions and speed
|
43 |
+
def compare_transcriptions(audio):
|
44 |
+
if audio is None:
|
45 |
+
raise gr.Error("No audio file submitted! Please record an audio before submitting your request.")
|
46 |
+
|
47 |
+
# Run both transcriptions in parallel
|
48 |
+
with concurrent.futures.ThreadPoolExecutor() as executor:
|
49 |
+
future_turbo = executor.submit(transcribe_turbo, audio)
|
50 |
+
future_standard = executor.submit(transcribe_standard, audio)
|
51 |
+
|
52 |
+
# Get the results
|
53 |
+
text_turbo, time_turbo = future_turbo.result()
|
54 |
+
text_standard, time_standard = future_standard.result()
|
55 |
+
|
56 |
+
# Return both transcriptions and processing times
|
57 |
+
return (text_standard, f"{time_standard:.2f} seconds"), (text_turbo, f"{time_turbo:.2f} seconds")
|
58 |
+
|
59 |
+
css = """
|
60 |
+
h1 {
|
61 |
+
text-align: center;
|
62 |
+
display:block;
|
63 |
+
}
|
64 |
+
"""
|
65 |
+
|
66 |
+
# Gradio Interface
|
67 |
+
with gr.Blocks(css=css) as demo:
|
68 |
+
# Title and description
|
69 |
+
gr.Markdown("# Whisper large-v3-turbo ...vs... Whisper large-v3")
|
70 |
+
gr.Markdown("This app compares the transcription performance and processing time between openAI 'Whisper large-v3' and 'Whisper large-v3-turbo' models")
|
71 |
+
|
72 |
+
with gr.Column():
|
73 |
+
with gr.Row():
|
74 |
+
with gr.Group():
|
75 |
+
audio_input = gr.Audio(sources=["microphone"], type="filepath")
|
76 |
+
transcribe_button = gr.Button("Start transcription", variant="primary")
|
77 |
+
|
78 |
+
with gr.Row():
|
79 |
+
with gr.Row():
|
80 |
+
with gr.Group():
|
81 |
+
gr.Markdown("### π **Standard model**")
|
82 |
+
standard_output = gr.Textbox(label="Transcription")
|
83 |
+
standard_time = gr.Textbox(label="Processing Time")
|
84 |
+
with gr.Group():
|
85 |
+
gr.Markdown("### β‘ **Turbo model**")
|
86 |
+
turbo_output = gr.Textbox(label="Transcription")
|
87 |
+
turbo_time = gr.Textbox(label="Processing Time")
|
88 |
+
|
89 |
+
# Set up the interaction
|
90 |
+
transcribe_button.click(fn=compare_transcriptions, inputs=audio_input, outputs=[standard_output, standard_time, turbo_output, turbo_time])
|
91 |
+
|
92 |
+
# Launch the demo
|
93 |
+
demo.launch()
|