Spaces:
Running
Running
Commit
•
a24b835
1
Parent(s):
3af9b1f
Update app.py
Browse files
app.py
CHANGED
@@ -6,7 +6,7 @@ text = "[Github]"
|
|
6 |
paper_link = "https://arxiv.org/pdf/2307.03183.pdf"
|
7 |
paper_text = "[Paper]"
|
8 |
|
9 |
-
model = whisper.load_model(
|
10 |
print('model loaded')
|
11 |
|
12 |
def round_time_resolution(time_resolution):
|
@@ -16,9 +16,11 @@ def round_time_resolution(time_resolution):
|
|
16 |
return rounded_time_resolution
|
17 |
|
18 |
def predict(audio_path_m, audio_path_t, time_resolution):
|
19 |
-
|
20 |
-
|
21 |
-
|
|
|
|
|
22 |
else:
|
23 |
audio_path = audio_path_m or audio_path_t
|
24 |
audio_tagging_time_resolution = round_time_resolution(time_resolution)
|
@@ -26,19 +28,19 @@ def predict(audio_path_m, audio_path_t, time_resolution):
|
|
26 |
audio_tag_result = whisper.parse_at_label(result, language='follow_asr', top_k=5, p_threshold=-1, include_class_list=list(range(527)))
|
27 |
asr_output = ""
|
28 |
for segment in result['segments']:
|
29 |
-
asr_output = asr_output +
|
30 |
at_output = ""
|
31 |
for segment in audio_tag_result:
|
32 |
print(segment)
|
33 |
-
at_output = at_output +
|
34 |
print(at_output)
|
35 |
return asr_output, at_output
|
36 |
|
37 |
iface = gr.Interface(fn=predict,
|
38 |
-
inputs=[gr.Audio(type="filepath", source='microphone'), gr.Audio(type="filepath"), gr.Textbox(value='10', label='Time Resolution in Seconds (Must be must be an integer multiple of 0.4, e.g., 0.4, 2, 10)')],
|
39 |
outputs=[gr.Textbox(label="Speech Output"), gr.Textbox(label="Audio Tag Output")],
|
40 |
cache_examples=True,
|
41 |
title="Quick Demo of Whisper-AT",
|
42 |
description="We are glad to introduce Whisper-AT - A new joint audio tagging and speech recognition model. It outputs background sound labels in addition to text." + f"<a href='{paper_link}'>{paper_text}</a> " + f"<a href='{link}'>{text}</a> <br>" +
|
43 |
-
"Whisper-AT is authored by Yuan Gong, Sameer Khurana, Leonid Karlinsky, and James Glass (MIT & MIT-IBM Watson AI Lab).")
|
44 |
iface.launch(debug=True)
|
|
|
6 |
paper_link = "https://arxiv.org/pdf/2307.03183.pdf"
|
7 |
paper_text = "[Paper]"
|
8 |
|
9 |
+
model = whisper.load_model('large-v1')
|
10 |
print('model loaded')
|
11 |
|
12 |
def round_time_resolution(time_resolution):
|
|
|
16 |
return rounded_time_resolution
|
17 |
|
18 |
def predict(audio_path_m, audio_path_t, time_resolution):
|
19 |
+
# print(audio_path_m, audio_path_t)
|
20 |
+
# print(type(audio_path_m), type(audio_path_t))
|
21 |
+
#return audio_path_m, audio_path_t
|
22 |
+
if ((audio_path_m is None) != (audio_path_t is None)) == False:
|
23 |
+
return "Please upload and only upload one recording, either upload the audio file or record using microphone.", "Please upload and only upload one recording, either upload the audio file or record using microphone."
|
24 |
else:
|
25 |
audio_path = audio_path_m or audio_path_t
|
26 |
audio_tagging_time_resolution = round_time_resolution(time_resolution)
|
|
|
28 |
audio_tag_result = whisper.parse_at_label(result, language='follow_asr', top_k=5, p_threshold=-1, include_class_list=list(range(527)))
|
29 |
asr_output = ""
|
30 |
for segment in result['segments']:
|
31 |
+
asr_output = asr_output + format(segment['start'], ".1f") + 's-' + format(segment['end'], ".1f") + 's: ' + segment['text'] + '\n'
|
32 |
at_output = ""
|
33 |
for segment in audio_tag_result:
|
34 |
print(segment)
|
35 |
+
at_output = at_output + format(segment['time']['start'], ".1f") + 's-' + format(segment['time']['end'], ".1f") + 's: ' + ','.join([x[0] for x in segment['audio tags']]) + '\n'
|
36 |
print(at_output)
|
37 |
return asr_output, at_output
|
38 |
|
39 |
iface = gr.Interface(fn=predict,
|
40 |
+
inputs=[gr.Audio(type="filepath", source='microphone', label='Please either upload an audio file or record using the microphone.', show_label=True), gr.Audio(type="filepath"), gr.Textbox(value='10', label='Time Resolution in Seconds (Must be must be an integer multiple of 0.4, e.g., 0.4, 2, 10)')],
|
41 |
outputs=[gr.Textbox(label="Speech Output"), gr.Textbox(label="Audio Tag Output")],
|
42 |
cache_examples=True,
|
43 |
title="Quick Demo of Whisper-AT",
|
44 |
description="We are glad to introduce Whisper-AT - A new joint audio tagging and speech recognition model. It outputs background sound labels in addition to text." + f"<a href='{paper_link}'>{paper_text}</a> " + f"<a href='{link}'>{text}</a> <br>" +
|
45 |
+
"Whisper-AT is authored by Yuan Gong, Sameer Khurana, Leonid Karlinsky, and James Glass (MIT & MIT-IBM Watson AI Lab). It is an Interspeech 2023 paper.")
|
46 |
iface.launch(debug=True)
|