add UI interface
Browse files
app.py
CHANGED
@@ -1,4 +1,4 @@
|
|
1 |
-
import whisper
|
2 |
import gradio as gr
|
3 |
import datetime
|
4 |
|
@@ -6,96 +6,144 @@ import subprocess
|
|
6 |
import wave
|
7 |
import contextlib
|
8 |
|
9 |
-
import torch
|
10 |
-
import pyannote.audio
|
11 |
-
from pyannote.audio.pipelines.speaker_verification import PretrainedSpeakerEmbedding
|
12 |
-
from pyannote.audio import Audio
|
13 |
-
from pyannote.core import Segment
|
14 |
-
from sklearn.cluster import AgglomerativeClustering
|
15 |
-
import numpy as np
|
16 |
-
|
17 |
-
model = whisper.load_model("large-v2")
|
18 |
-
embedding_model = PretrainedSpeakerEmbedding(
|
19 |
-
|
20 |
-
|
21 |
-
)
|
22 |
-
|
23 |
-
def transcribe(audio, num_speakers):
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
def convert_to_wav(path):
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
def get_duration(path):
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
def make_embeddings(path, segments, duration):
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
audio = Audio()
|
67 |
-
|
68 |
-
def segment_embedding(path, segment, duration):
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
def add_speaker_labels(segments, embeddings, num_speakers):
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
def time(secs):
|
83 |
-
|
84 |
-
|
85 |
-
def get_output(segments):
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
95 |
|
96 |
with gr.Blocks() as demo:
|
97 |
-
with gr.
|
98 |
-
gr.
|
99 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
100 |
|
101 |
demo.launch()
|
|
|
1 |
+
# import whisper
|
2 |
import gradio as gr
|
3 |
import datetime
|
4 |
|
|
|
6 |
import wave
|
7 |
import contextlib
|
8 |
|
9 |
+
# import torch
|
10 |
+
# import pyannote.audio
|
11 |
+
# from pyannote.audio.pipelines.speaker_verification import PretrainedSpeakerEmbedding
|
12 |
+
# from pyannote.audio import Audio
|
13 |
+
# from pyannote.core import Segment
|
14 |
+
# from sklearn.cluster import AgglomerativeClustering
|
15 |
+
# import numpy as np
|
16 |
+
|
17 |
+
# model = whisper.load_model("large-v2")
|
18 |
+
# embedding_model = PretrainedSpeakerEmbedding(
|
19 |
+
# "speechbrain/spkrec-ecapa-voxceleb",
|
20 |
+
# device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
21 |
+
# )
|
22 |
+
|
23 |
+
# def transcribe(audio, num_speakers):
|
24 |
+
# path, error = convert_to_wav(audio)
|
25 |
+
# if error is not None:
|
26 |
+
# return error
|
27 |
+
|
28 |
+
# duration = get_duration(path)
|
29 |
+
# if duration > 4 * 60 * 60:
|
30 |
+
# return "Audio duration too long"
|
31 |
+
|
32 |
+
# result = model.transcribe(path)
|
33 |
+
# segments = result["segments"]
|
34 |
+
|
35 |
+
# num_speakers = min(max(round(num_speakers), 1), len(segments))
|
36 |
+
# if len(segments) == 1:
|
37 |
+
# segments[0]['speaker'] = 'SPEAKER 1'
|
38 |
+
# else:
|
39 |
+
# embeddings = make_embeddings(path, segments, duration)
|
40 |
+
# add_speaker_labels(segments, embeddings, num_speakers)
|
41 |
+
# output = get_output(segments)
|
42 |
+
# return output
|
43 |
+
|
44 |
+
# def convert_to_wav(path):
|
45 |
+
# if path[-3:] != 'wav':
|
46 |
+
# new_path = '.'.join(path.split('.')[:-1]) + '.wav'
|
47 |
+
# try:
|
48 |
+
# subprocess.call(['ffmpeg', '-i', path, new_path, '-y'])
|
49 |
+
# except:
|
50 |
+
# return path, 'Error: Could not convert file to .wav'
|
51 |
+
# path = new_path
|
52 |
+
# return path, None
|
53 |
+
|
54 |
+
# def get_duration(path):
|
55 |
+
# with contextlib.closing(wave.open(path,'r')) as f:
|
56 |
+
# frames = f.getnframes()
|
57 |
+
# rate = f.getframerate()
|
58 |
+
# return frames / float(rate)
|
59 |
+
|
60 |
+
# def make_embeddings(path, segments, duration):
|
61 |
+
# embeddings = np.zeros(shape=(len(segments), 192))
|
62 |
+
# for i, segment in enumerate(segments):
|
63 |
+
# embeddings[i] = segment_embedding(path, segment, duration)
|
64 |
+
# return np.nan_to_num(embeddings)
|
65 |
+
|
66 |
+
# audio = Audio()
|
67 |
+
|
68 |
+
# def segment_embedding(path, segment, duration):
|
69 |
+
# start = segment["start"]
|
70 |
+
# # Whisper overshoots the end timestamp in the last segment
|
71 |
+
# end = min(duration, segment["end"])
|
72 |
+
# clip = Segment(start, end)
|
73 |
+
# waveform, sample_rate = audio.crop(path, clip)
|
74 |
+
# return embedding_model(waveform[None])
|
75 |
+
|
76 |
+
# def add_speaker_labels(segments, embeddings, num_speakers):
|
77 |
+
# clustering = AgglomerativeClustering(num_speakers).fit(embeddings)
|
78 |
+
# labels = clustering.labels_
|
79 |
+
# for i in range(len(segments)):
|
80 |
+
# segments[i]["speaker"] = 'SPEAKER ' + str(labels[i] + 1)
|
81 |
+
|
82 |
+
# def time(secs):
|
83 |
+
# return datetime.timedelta(seconds=round(secs))
|
84 |
+
|
85 |
+
# def get_output(segments):
|
86 |
+
# output = ''
|
87 |
+
# for (i, segment) in enumerate(segments):
|
88 |
+
# if i == 0 or segments[i - 1]["speaker"] != segment["speaker"]:
|
89 |
+
# if i != 0:
|
90 |
+
# output += '\n\n'
|
91 |
+
# output += segment["speaker"] + ' ' + str(time(segment["start"])) + '\n\n'
|
92 |
+
# output += segment["text"][1:] + ' '
|
93 |
+
# return output
|
94 |
+
|
95 |
+
s = ""
|
96 |
+
|
97 |
+
def greet1(name):
|
98 |
+
global s
|
99 |
+
s = "modified"
|
100 |
+
return "Hello " + name + "!"
|
101 |
+
|
102 |
+
|
103 |
+
def greet2(name):
|
104 |
+
return "Hi " + name + "!" + " " + s
|
105 |
+
|
106 |
+
|
107 |
+
def greet3(name):
|
108 |
+
return "Hola " + name + "!"
|
109 |
|
110 |
with gr.Blocks() as demo:
|
111 |
+
with gr.Row():
|
112 |
+
with gr.Column():
|
113 |
+
audio_file = gr.UploadButton(label="Upload a Audio file (.wav)")
|
114 |
+
# name = gr.Textbox(label="Name", placeholder="Name") # TODO: remove
|
115 |
+
number_of_speakers = gr.Number(label="Number of Speakers", value=2)
|
116 |
+
with gr.Row():
|
117 |
+
btn_clear = gr.Button(value="Clear")
|
118 |
+
btn_submit = gr.Button(value="Submit")
|
119 |
+
with gr.Column():
|
120 |
+
title = gr.Textbox(label="Title", placeholder="Title for Conversation")
|
121 |
+
short_summary = gr.Textbox(label="Short Summary", placeholder="Short Summary for Conversation")
|
122 |
+
sentiment_analysis = gr.Textbox(label="Sentiment Analysis", placeholder="Sentiment Analysis for Conversation")
|
123 |
+
quality = gr.Textbox(label="Quality of Conversation", placeholder="Quality of Conversation")
|
124 |
+
detailed_summary = gr.Textbox(label="Detailed Summary", placeholder="Detailed Summary for Conversation")
|
125 |
+
gr.Markdown("## Examples")
|
126 |
+
gr.Examples(
|
127 |
+
examples=[
|
128 |
+
[
|
129 |
+
"Harsh",
|
130 |
+
2,
|
131 |
+
],
|
132 |
+
[
|
133 |
+
"Rahul",
|
134 |
+
2,
|
135 |
+
],
|
136 |
+
],
|
137 |
+
inputs=[title],
|
138 |
+
outputs=[short_summary],
|
139 |
+
fn=greet1,
|
140 |
+
cache_examples=True,
|
141 |
+
)
|
142 |
+
gr.Markdown(
|
143 |
+
"""
|
144 |
+
See [github.com/facebookresearch/audiocraft](https://github.com/facebookresearch/audiocraft)
|
145 |
+
for more details.
|
146 |
+
"""
|
147 |
+
)
|
148 |
|
149 |
demo.launch()
|