Update app.py
Browse filesstreaming output + supporting long video
app.py
CHANGED
@@ -2,9 +2,45 @@ import gradio as gr
|
|
2 |
import openai
|
3 |
import yt_dlp
|
4 |
import os
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
openai.api_key = os.environ['OPENAI_API_KEY']
|
6 |
|
7 |
def asr(url):
|
|
|
|
|
8 |
# download audio
|
9 |
# Options for youtube-dl
|
10 |
ydl_opts = {
|
@@ -22,21 +58,49 @@ def asr(url):
|
|
22 |
audio_file_name = "audio_downloaded.{}".format(info_dict["ext"])
|
23 |
else:
|
24 |
return "下载音频发生错误,请确认链接再试一次。", "Error downloading the audio. Check the URL and try again."
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
|
|
|
|
|
|
36 |
|
|
|
|
|
|
|
|
|
|
|
37 |
# delete the video
|
38 |
os.system("rm {}".format(audio_file_name))
|
39 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
|
41 |
title = """
|
42 |
轻声细译"""
|
@@ -46,7 +110,6 @@ instruction = """
|
|
46 |
一键输入视频链接,轻松实现中文翻译,畅享视频无障碍沟通 <span style="color: grey;">-- powered by OpenAI Whisper & ChatGPT.</span>.<br>
|
47 |
|
48 |
1.将视频链接(支持Twitter、YouTube)复制粘贴至输入框,点击提交(Submit)即可;
|
49 |
-
2.为保证翻译质量,目前仅支持处理时长不超过5分钟的短视频。
|
50 |
</div>"""
|
51 |
# Create a text input component
|
52 |
text_input = gr.inputs.Textbox()
|
@@ -58,6 +121,6 @@ demo = gr.Interface(fn=asr,
|
|
58 |
gr.outputs.Textbox(label="英文")
|
59 |
],
|
60 |
title=title,
|
61 |
-
description=instruction,theme='huggingface'
|
62 |
-
|
63 |
demo.launch()
|
|
|
2 |
import openai
|
3 |
import yt_dlp
|
4 |
import os
|
5 |
+
import io
|
6 |
+
import tempfile
|
7 |
+
from pydub import AudioSegment
|
8 |
+
|
9 |
+
def split_audio(file_path, chunk_length_ms):
|
10 |
+
audio = AudioSegment.from_file(file_path)
|
11 |
+
duration = len(audio)
|
12 |
+
chunks = []
|
13 |
+
start_time = 0
|
14 |
+
while start_time < duration:
|
15 |
+
end_time = start_time + chunk_length_ms
|
16 |
+
if end_time > duration:
|
17 |
+
end_time = duration
|
18 |
+
chunk = audio[start_time:end_time]
|
19 |
+
chunks.append(chunk)
|
20 |
+
start_time += chunk_length_ms
|
21 |
+
return chunks
|
22 |
+
|
23 |
+
def split_string_by_tokens(text, max_tokens=500):
|
24 |
+
words = text.split()
|
25 |
+
chunks = []
|
26 |
+
current_chunk = []
|
27 |
+
|
28 |
+
for word in words:
|
29 |
+
current_chunk.append(word)
|
30 |
+
if len(current_chunk) >= max_tokens:
|
31 |
+
chunks.append(' '.join(current_chunk))
|
32 |
+
current_chunk = []
|
33 |
+
|
34 |
+
if current_chunk:
|
35 |
+
chunks.append(' '.join(current_chunk))
|
36 |
+
|
37 |
+
return chunks
|
38 |
+
|
39 |
openai.api_key = os.environ['OPENAI_API_KEY']
|
40 |
|
41 |
def asr(url):
|
42 |
+
# delete the video
|
43 |
+
os.system("rm *audio_download*")
|
44 |
# download audio
|
45 |
# Options for youtube-dl
|
46 |
ydl_opts = {
|
|
|
58 |
audio_file_name = "audio_downloaded.{}".format(info_dict["ext"])
|
59 |
else:
|
60 |
return "下载音频发生错误,请确认链接再试一次。", "Error downloading the audio. Check the URL and try again."
|
61 |
+
|
62 |
+
yield "下载视频完成. 开始分割视频...", ""
|
63 |
+
chunks = split_audio(audio_file_name, chunk_length_ms=30 * 1000)
|
64 |
+
transcripts = []
|
65 |
+
|
66 |
+
for idx, chunk in enumerate(chunks):
|
67 |
+
|
68 |
+
temp_file_path = None
|
69 |
+
with tempfile.NamedTemporaryFile(mode="wb", suffix=".wav", delete=False) as temp_file:
|
70 |
+
temp_file_path = temp_file.name
|
71 |
+
chunk.export(temp_file.name, format="wav")
|
72 |
+
|
73 |
+
with open(temp_file_path, "rb") as temp_file:
|
74 |
+
transcript = openai.Audio.transcribe("whisper-1", temp_file)
|
75 |
|
76 |
+
os.remove(temp_file_path)
|
77 |
+
transcripts.append(transcript["text"])
|
78 |
+
|
79 |
+
yield "请耐心等待语音识别完成...({}/{})".format(idx + 1, len(chunks)), " ".join(transcripts)
|
80 |
+
|
81 |
# delete the video
|
82 |
os.system("rm {}".format(audio_file_name))
|
83 |
+
|
84 |
+
translations = []
|
85 |
+
full_transcript = " ".join(transcripts)
|
86 |
+
# split into 500 tokens
|
87 |
+
transcript_chunks = split_string_by_tokens(full_transcript, max_tokens=500)
|
88 |
+
yield "语音识别完成, 开始翻译...(0/{})".format(len(transcript_chunks)), full_transcript
|
89 |
+
# split transcripts if its too long
|
90 |
+
for idx, transcript in enumerate(transcript_chunks):
|
91 |
+
output = openai.ChatCompletion.create(
|
92 |
+
model="gpt-3.5-turbo",
|
93 |
+
messages=[
|
94 |
+
{"role": "user", "content": "Transcript: {transcript}. \n Translate the video conversation transcript into fluent Chinese. Chinese: ".format(transcript=transcript)},
|
95 |
+
]
|
96 |
+
)
|
97 |
+
translation = output['choices'][0]['message']['content']
|
98 |
+
translations.append(translation)
|
99 |
+
|
100 |
+
yield "请耐心等候翻译:({}/{})...".format(idx+1, len(transcript_chunks)) + " ".join(translations), " ".join(transcripts)
|
101 |
+
|
102 |
+
full_translation = " ".join(translations)
|
103 |
+
yield full_translation, full_transcript
|
104 |
|
105 |
title = """
|
106 |
轻声细译"""
|
|
|
110 |
一键输入视频链接,轻松实现中文翻译,畅享视频无障碍沟通 <span style="color: grey;">-- powered by OpenAI Whisper & ChatGPT.</span>.<br>
|
111 |
|
112 |
1.将视频链接(支持Twitter、YouTube)复制粘贴至输入框,点击提交(Submit)即可;
|
|
|
113 |
</div>"""
|
114 |
# Create a text input component
|
115 |
text_input = gr.inputs.Textbox()
|
|
|
121 |
gr.outputs.Textbox(label="英文")
|
122 |
],
|
123 |
title=title,
|
124 |
+
description=instruction,theme='huggingface')
|
125 |
+
demo.queue()
|
126 |
demo.launch()
|