Spaces:
Runtime error
Runtime error
Davidsamuel101
commited on
Commit
•
0ab122b
1
Parent(s):
a77bcb7
Initial Commit
Browse files- app.py +113 -0
- requirements.txt +1 -0
app.py
ADDED
@@ -0,0 +1,113 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from speechline.transcribers import Wav2Vec2Transcriber
|
2 |
+
from speechline.segmenters import SilenceSegmenter, WordOverlapSegmenter
|
3 |
+
from speechline.utils.tokenizer import WordTokenizer
|
4 |
+
from datasets import Dataset, Audio
|
5 |
+
from pathlib import Path
|
6 |
+
|
7 |
+
import gradio as gr
|
8 |
+
import shutil
|
9 |
+
|
10 |
+
max_textboxes=5
|
11 |
+
|
12 |
+
def preprocess(audio_path, transcriber):
|
13 |
+
dataset = Dataset.from_dict({"audio": [audio_path]})
|
14 |
+
dataset = dataset.cast_column("audio", Audio(sampling_rate=transcriber.sampling_rate))
|
15 |
+
return dataset
|
16 |
+
|
17 |
+
def transcribe(audio_path, transcriber):
|
18 |
+
dataset = preprocess(audio_path, transcriber)
|
19 |
+
output_offsets = transcriber.predict(dataset, output_offsets=True)
|
20 |
+
return output_offsets
|
21 |
+
|
22 |
+
def segmentation_interface(choice):
|
23 |
+
if choice == "silence":
|
24 |
+
return gr.update(visible=True), gr.update(visible=False)
|
25 |
+
elif choice == "word_overlap":
|
26 |
+
return gr.update(visible=False), gr.update(visible=True)
|
27 |
+
else:
|
28 |
+
return gr.update(visible=False), gr.update(visible=False)
|
29 |
+
|
30 |
+
|
31 |
+
def process(audio_path, model, segmentation_type, silence_duration, ground_truth):
|
32 |
+
output_dir = "./audio_chunks"
|
33 |
+
|
34 |
+
transcriber = Wav2Vec2Transcriber(model)
|
35 |
+
output_offsets = transcribe(audio_path, transcriber)
|
36 |
+
|
37 |
+
if segmentation_type == "silence":
|
38 |
+
segmenter = SilenceSegmenter()
|
39 |
+
elif segmentation_type == "word_overlap":
|
40 |
+
segmenter = WordOverlapSegmenter()
|
41 |
+
|
42 |
+
tokenizer = WordTokenizer()
|
43 |
+
|
44 |
+
if os.path.exists(f"{output_dir}/tmp"):
|
45 |
+
shutil.rmtree(f"{output_dir}/tmp")
|
46 |
+
|
47 |
+
segmenter.chunk_audio_segments(
|
48 |
+
audio_path,
|
49 |
+
output_dir,
|
50 |
+
output_offsets[0],
|
51 |
+
minimum_chunk_duration=0,
|
52 |
+
silence_duration=silence_duration,
|
53 |
+
ground_truth=tokenizer(ground_truth),
|
54 |
+
)
|
55 |
+
|
56 |
+
outputs = []
|
57 |
+
idx = 0
|
58 |
+
for path in sorted(Path(f"{output_dir}/tmp").iterdir()):
|
59 |
+
if str(path).split('.')[-1] == 'tsv':
|
60 |
+
gt = pd.read_csv(path, sep='\t', names=["start_offset", "end_offset", "text"])
|
61 |
+
outputs.append(gr.Dataframe.update(value=gt,visible=True))
|
62 |
+
idx+=1
|
63 |
+
if str(path).split('.')[-1] == 'wav':
|
64 |
+
audio = (str(path))
|
65 |
+
outputs.append(gr.Audio.update(value=audio, visible=True))
|
66 |
+
|
67 |
+
for i in range(max_textboxes-idx):
|
68 |
+
outputs.append(gr.Dataframe.update(visible=False))
|
69 |
+
outputs.append(gr.Audio.update(visible=False))
|
70 |
+
outputs.append(gr.Column.update(visible=True))
|
71 |
+
return outputs
|
72 |
+
|
73 |
+
with gr.Blocks() as demo:
|
74 |
+
with gr.Row():
|
75 |
+
with gr.Column():
|
76 |
+
audio = gr.Audio(type="filepath")
|
77 |
+
radio = gr.Radio(["silence", "word_overlap"], label="Select Segmentation Method", required=True)
|
78 |
+
model = gr.Dropdown(["facebook/wav2vec2-base-960h", "bookbot/wav2vec-en", "bookbot/wav2vec-id"], value="facebook/wav2vec2-base-960h", label="Select Model")
|
79 |
+
slider = gr.Slider(0, 100, value=3, step=0.1, visible=False)
|
80 |
+
gt = gr.Textbox(label="Ground Truth", placeholder="Enter Ground Truth Text", interactive=True, visible=False)
|
81 |
+
|
82 |
+
radio.change(fn=segmentation_interface, inputs=radio, outputs=[slider, gt])
|
83 |
+
|
84 |
+
inputs = [audio, model, radio, slider, gt]
|
85 |
+
transcribe_btn = gr.Button("Transcribe")
|
86 |
+
|
87 |
+
|
88 |
+
with gr.Column(visible=False) as output_col:
|
89 |
+
outputs = []
|
90 |
+
gt1 = gr.Dataframe(visible=False)
|
91 |
+
audio1 = gr.Audio(visible=False)
|
92 |
+
|
93 |
+
gt2 = gr.Dataframe(visible=False)
|
94 |
+
audio2 = gr.Audio(visible=False)
|
95 |
+
|
96 |
+
gt3 = gr.Dataframe(visible=False)
|
97 |
+
audio3 = gr.Audio(visible=False)
|
98 |
+
|
99 |
+
gt4 = gr.Dataframe(visible=False)
|
100 |
+
audio4 = gr.Audio(visible=False)
|
101 |
+
|
102 |
+
gt5 = gr.Dataframe(visible=False)
|
103 |
+
audio5 = gr.Audio(visible=False)
|
104 |
+
|
105 |
+
|
106 |
+
for i in range(max_textboxes):
|
107 |
+
outputs.append(gr.Dataframe(visible=False))
|
108 |
+
outputs.append(gr.Audio(visible=False))
|
109 |
+
outputs.append(output_col)
|
110 |
+
|
111 |
+
transcribe_btn.click(fn=process, inputs=inputs, outputs=outputs)
|
112 |
+
|
113 |
+
demo.queue().launch()
|
requirements.txt
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
speechline @ git+https://github.com/bookbot-kids/speechline.git
|