Spaces:
Build error
Build error
app.py
CHANGED
@@ -1 +1,115 @@
|
|
1 |
-
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
|
3 |
+
#final
|
4 |
+
import gradio as gr
|
5 |
+
#import json
|
6 |
+
#from difflib import Differ
|
7 |
+
import ffmpeg
|
8 |
+
#import os
|
9 |
+
from pathlib import Path
|
10 |
+
#import time
|
11 |
+
|
12 |
+
API_URL = "https://api-inference.huggingface.co/models/facebook/wav2vec2-base-960h"
|
13 |
+
headers = {"Authorization": "Bearer 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"}
|
14 |
+
|
15 |
+
#convert video to audio
|
16 |
+
video_path = Path("/content/gdrive/My Drive/AI/videoedit/ShiaLaBeouf.mp4")
|
17 |
+
audio_memory, _ = ffmpeg.input(video_path).output('-', format="wav", ac=1, ar='16k').overwrite_output().global_args('-loglevel', 'quiet').run(capture_stdout=True)
|
18 |
+
|
19 |
+
#calling the hosted model
|
20 |
+
def query_api(audio_bytes: bytes):
|
21 |
+
"""
|
22 |
+
Query for Huggingface Inference API for Automatic Speech Recognition task
|
23 |
+
"""
|
24 |
+
payload = json.dumps({
|
25 |
+
"inputs": base64.b64encode(audio_bytes).decode("utf-8"),
|
26 |
+
"parameters": {
|
27 |
+
"return_timestamps": "char",
|
28 |
+
"chunk_length_s": 10,
|
29 |
+
"stride_length_s": [4, 2]
|
30 |
+
},
|
31 |
+
"options": {"use_gpu": False}
|
32 |
+
}).encode("utf-8")
|
33 |
+
|
34 |
+
response = requests.request(
|
35 |
+
"POST", API_URL, headers=headers, data=payload)
|
36 |
+
json_reponse = json.loads(response.content.decode("utf-8"))
|
37 |
+
return json_reponse
|
38 |
+
|
39 |
+
#Getting transcripts using wav2Vec2 huggingface hosted accelerated inference
|
40 |
+
#sending audio file in request along with stride and chunk length information
|
41 |
+
model_response = query_api(audio_memory)
|
42 |
+
|
43 |
+
#model response has both - transcripts as well as character timestamps or chunks
|
44 |
+
transcription = model_response["text"].lower()
|
45 |
+
chnk = model_response["chunks"]
|
46 |
+
|
47 |
+
#creating lists from chunks to consume downstream easily
|
48 |
+
timestamps = [[chunk["text"].lower(), chunk["timestamp"][0], chunk["timestamp"][1]]
|
49 |
+
for chunk in chnk]
|
50 |
+
|
51 |
+
|
52 |
+
#getting word timestams from character timestamps
|
53 |
+
def get_word_timestamps(timestamps):
|
54 |
+
words, word = [], []
|
55 |
+
letter_timestamp, word_timestamp, words_timestamp = [], [], []
|
56 |
+
for idx,entry in enumerate(timestamps):
|
57 |
+
word.append(entry[0])
|
58 |
+
letter_timestamp.append(entry[1])
|
59 |
+
if entry[0] == ' ':
|
60 |
+
words.append(''.join(word))
|
61 |
+
word_timestamp.append(letter_timestamp[0])
|
62 |
+
word_timestamp.append(timestamps[idx-1][2])
|
63 |
+
words_timestamp.append(word_timestamp)
|
64 |
+
word, word_timestamp, letter_timestamp = [], [], []
|
65 |
+
|
66 |
+
words = [word.strip() for word in words]
|
67 |
+
return words, words_timestamp
|
68 |
+
|
69 |
+
words, words_timestamp = get_word_timestamps(timestamps)
|
70 |
+
#words = [word.strip() for word in words]
|
71 |
+
|
72 |
+
print(f"Total words in the audio transcript is:{len(words)}, transcript word list is :{words}")
|
73 |
+
print(f"Total Word timestamps derived fromcharacter timestamp are :{len(words_timestamp)}, Word timestamps are :{words_timestamp}")
|
74 |
+
|
75 |
+
#creating list from input gif transcript
|
76 |
+
gif = "don't let your dreams be dreams"
|
77 |
+
giflist = gif.split()
|
78 |
+
|
79 |
+
#getting index of gif words in main transcript
|
80 |
+
def get_gif_word_indexes(total_words_list, gif_words_list):
|
81 |
+
if not gif_words_list:
|
82 |
+
return
|
83 |
+
# just optimization
|
84 |
+
lengthgif_words_list = len(gif_words_list)
|
85 |
+
firstgif_words_list = gif_words_list[0]
|
86 |
+
for idx, item in enumerate(total_words_list):
|
87 |
+
if item == firstgif_words_list:
|
88 |
+
if total_words_list[idx:idx+lengthgif_words_list] == gif_words_list:
|
89 |
+
yield tuple(range(idx, idx+lengthgif_words_list))
|
90 |
+
|
91 |
+
#getting gif indexes from the generator
|
92 |
+
giflist_indxs = list(list(get_gif_word_indexes(words, giflist))[0])
|
93 |
+
|
94 |
+
#getting start and end timestamps for gif transcript
|
95 |
+
def get_gif_timestamps(giflist_indxs):
|
96 |
+
#giflist_indxs = list(list(get_gif_word_indexes(words, giflist))[0])
|
97 |
+
min_idx = min(giflist_indxs)
|
98 |
+
max_idx = max(giflist_indxs)
|
99 |
+
|
100 |
+
gif_words_timestamp = words_timestamp[min_idx : max_idx+1]
|
101 |
+
start_seconds, end_seconds = gif_words_timestamp[0][0], gif_words_timestamp[-1][-1]
|
102 |
+
return start_seconds, end_seconds
|
103 |
+
|
104 |
+
#getting start and end timestamps for a gif video
|
105 |
+
start_seconds, end_seconds = get_gif_timestamps(giflist_indxs)
|
106 |
+
|
107 |
+
#extracting the video and building and serving a .gif image
|
108 |
+
def generate_gif(start_seconds, end_seconds):
|
109 |
+
final_clip = video.subclip(start_seconds, end_seconds)
|
110 |
+
#final_clip.write_videofile("/content/gdrive/My Drive/AI/videoedit/gif1.mp4")
|
111 |
+
final_clip.write_gif("/content/gdrive/My Drive/AI/videoedit/gif1.gif",)
|
112 |
+
final_clip.close()
|
113 |
+
return
|
114 |
+
|
115 |
+
generate_gif(start_seconds, end_seconds)
|