2022-07-08
Cut Music Scenes With Lyrics And Bpm

Cut Music Segments With Lyrics and BPM

def compare(a,b,reverse=False):

seg_low, seg_high = get_allowed_segments(bpm, low, high, tolerance=0.8) # the tolerance is compared with a common function called compare. it can be customized to output only value >=1 or vice versa.

candidates = sorted_lyrics_nearby_bpm_candidates + sorted_remained_bpm_candidates # priortize lyrics candidates.

Read More

2022-06-03
Neo4J Learning Notes

neo4j desktop provide free multi graph databases support, enable you to create and use multiple graph datases at once

to query undirected relationships:

match () – (p) return p

create fulltext index

create fulltext index lucene for (n:Person) on each [n.title, n.description]

call db.index.fulltext.queryNodes(“titlesAndDescriptions”, “Full Metal Jacket”) yield node, score return node, score

calculate customer rating cosine similarity for recommendation:

https://neo4j.com/graphgists/northwind-recommendation-engine

match (c1:customer)-[r1:rated]-(:product)-[r2:rated]-(c2:customer)

with sum(r1.score*r2.score) as dot_product,

sqrt(reduce(x=0, a in r1.score | x+a^2)) as r1_length,

sqrt(reduce(y=0, b in r2.score | y+b^2)) as r2_length

merge (c1)-[s:similarity}]-(c2)

set s += {score:dot_product/(r1_length*r2_length)

use collect to turn maps into lists:

match (p) return collect(p.names)

exist subquery:

match (n:Person) where exists {

match (n) –(t:Tech)

where size((t)-[:likes]-(:Person)) >2

}

return n.name

list comprehension:

return [x in range(0,10) where x%3 = 0| x/2] as list

return [x in range(0,10) where not x in range(4,10) |x ] as list

to use conditional matches or regular expressions:

match () – (p) where p.name in [“helen”] or p.name =~ “.chinese.“ return p

create index on properties:

create index for (n:Category) on (n.categoryName)

asterisks:

load csv:

load csv with headers from “http://localhost/person.csv“ as line

call {with line

merge (n:person {id: toInteger(line.id)})

set n.name = line.name

} in transactions of 2 rows

count nodes:

match (n) return count(n)

match relationship patterns:

match (n) -[:friend|hater*3]->(p) return p limit 20

node can have multiple labels, while relationship can only have one type, both specified after the colon.

simple case expression:

match(n)

return

case n.eyes

when “blue” then 1

when “brown” then 2

else 3

end as result

generic case expression

match (n)

return

case

when n.eyes = “blue” then 1

when n.age > 40 then 2

else 3 // if without else then return null

end as result

inequality symbol: <>

mutating updating node properties:

match(n)

set n+={name:”helen”} // if using = the properties will be totally replaced instead of update.

return n.name

merge can only ensure the existance of one node or pattern at a time, no comma

plus can concatenate strings

tenporal dataformat can be used as numbers to compute.

Map operators

. for static value access by key, [] for dynamic value access by key

List operators

  • for concatenation, IN to check existence of an element in a list, [] for accessing element(s) dynamically

recommendation steps:

first find targets by meta relatonships

next sort recommendation by frequency, ratings or occurance

third filter items by topics or properties

Read More

2022-05-28
Im Mitm 聊天软件Mitm

IM MITM 聊天软件 MITM

better do this in virtual enviorment without using any real world platform, just your own IM enviorment like a self-hosted IRC or something.

is there any existing solution like telegram-mitm or twitter mitm?

lua twitter automation, found on luarocks:

https://github.com/leafo/lua-twitter

scraper of tumblr, pinterest, youtube, reddit using api:

https://github.com/ScriptSmith/socialreaper

youtube search and youtube comment scraper

https://github.com/alexmercerind/youtube-search-python

https://github.com/egbertbouman/youtube-comment-downloader

youtube, youtube transcribe and youtube music api

https://github.com/srcecde/python-youtube-api

https://github.com/sigma67/ytmusicapi

https://github.com/jdepoix/youtube-transcript-api

https://github.com/youtube/api-samples

reddit scraper and analyzer

https://github.com/casperbh96/Web-Scraping-Reddit

https://github.com/umitkaanusta/reddit-detective

reddit api

https://github.com/praw-dev/praw

tumblr api

https://github.com/tumblr/pytumblr

tumblr scraper

https://github.com/henan715/tumblrScrapy

discord bot api:

https://github.com/discordjs/discord.js

twitter api

https://github.com/python-twitter-tools/twitter

twitter scraper

https://github.com/bisguzar/twitter-scraper

facebook api:

https://github.com/Schmavery/facebook-chat-api

facebook scraper:

https://github.com/kevinzg/facebook-scraper

instagram api:

https://github.com/facebookarchive/python-instagram

instagram scraper:

https://github.com/huaying/instagram-crawler

topic analysis among recent frequent conversations

procedures:

1.add two friends (active) and bridge them

2.intercept them, filter insecure data like screenshots, identities and explicit contents, and analyze needs (probably with your generated response)?

3.send intentional Ads and fix the conversation in three sentences.

Read More

2022-05-28
变声软件 Morphvox Alternatives

Real Time Voice Cloning by CorentinJ

感谢关注,UP在B站制作各种AI变声器模型。

AI白菜原创变声器软件下载链接:

https://huggingface.co/lj1995/VoiceConversionWebUI/resolve/main/vits_vc_gpu.7z

有超级会员的建议下度盘

https://pan.baidu.com/s/1EyM7SzFUpxlemJlrqaJbdQ

提取码j24b

已开源的模型在线使用链接(很久以前了):

https://huggingface.co/spaces/innnky/trump

https://huggingface.co/spaces/innnky/nanami

白嫖微软azure tts

API下线,原因见动态。

推理脚本

声线转换:https://colab.research.google.com/drive/1W6aoDMuTku8EDTuH7-okVaj-6kzbXz1m

TTS:https://colab.research.google.com/drive/1kpHzOHfWqM4pXxUiqxON9SvDTOeXGNI1

声线转换模型:https://obs.baimianxiao.cn/share/obs/sankagenkeshi/G_1293000.pth

TTS模型(多人物):https://obs.baimianxiao.cn/share/obs/sankagenkeshi/G_809000.pth

VITS仓库地址 https://github.com/jaywalnut310/vits/

本模型迭代1300 Epochs

使用2xA100训练六天,效果还可以。

感谢国家超级计算广州中心提供的算力支持。

猫雷唱歌

在线使用demo(45秒限制):https://huggingface.co/spaces/innnky/nyaru-svc2.0

模型github链接:https://github.com/innnky/so-vits-svc

自己训练数据集、离线使用、一键训练、一键合成:https://github.com/IceKyrin/sovits_guide

本模型训练使用数据集:猫雷歌回1小时+杂谈回6小时;opencpop5小时 虚拟歌手云灏 半小时

moegoe

多种语言混合tts

multilingual tts dl models

项目地址:https://github.com/luoyily/MoeTTS

MoeTTS V1.2.1 Update

适配Windows DPI缩放

加入中文g2p工具

加入FFmepg 音频转换工具

支持批量合成

支持自定义文件名

支持VITS语速调节

修复日语g2p gbk错误

支持主题切换

GUI设计优化:

弃用原项目中的text模块,重写了text to sequence,使用无需再替换symbols,无需替换cleaners 实现模型解压即用

加入专用配置文件,目前仅用于指定模型symbols

(另外我这莫得中文模型,去隔壁CjangCjengh那边薅了一个来为视频配音

基于StarGANv2-VC的声音转换模型

声音质量当然赶不上VITS这种基于TTS的VC,但效果已经挺不错的了。转换后的语音噪音大应该不是StarGAN的问题i,而是

后面接的vocoder(用的ParallelWaveGAN)。本来想使用HiFi-GAN的,但找到的预训练模型都和StarGAN不适配,我也不想

自己从头训一个HiFi-GAN……

Colab Demo:

https://colab.research.google.com/drive/1Xpn9yKBuJD59llXNJOrdUpFuiQNkwDqo?usp=sharing

Github:

https://github.com/Francis-Komizu/StarGANv2-VC

vits可以模仿声优的声线

softvc vits联合模型

Colab demo:

https://colab.research.google.com/drive/1OjfH2zpRkLFRp92aU6jAGhqZNopfZMjC?usp=sharing

项目代码:

https://github.com/Francis-Komizu/Sovits

歌曲变声 男的需要提高八度再变声 softvc Vits

猫雷歌曲变声器 singingvoiceconversion

派蒙在COLAB上面训练的notebook:

派蒙语音合成地址:

https://colab.research.google.com/drive/1HDV84t3N-yUEBXN8dDIDSv6CzEJykCLw#scrollTo=oiPvCIJ_MHot

请注意不要输入英文标点符号(忘记改了)[脱单doge]。切勿商用或进行18+、暴力、血腥内容传播

这个人还在训练其他人物的语音

b站有这些插件如何下载的教程

宿主:Studio One 5

降噪插件:RX系列Voice De-noise(可替换)

变声插件:LittleAlterBoy(不可替换)(是目前变声插件的最优选择,实在不行,老版本的Auto Tune也可以)

压缩器、去齿音:肥波系列Pro-C2,Pro-DS(可替换)

EQ均衡器:肥波系列Pro-Q3(可替换)

(再推荐一个soothe2,简单说这个插件可以让声音更自然一些,但是比较贵,可以搜索教学视频了解一下。)

最后,有问题可以在评论区,私信问我,看到的话会给予解答。

视频内BGM:

Fluffing a Duck–Kevin MacLeod

ひとときの安らぎ–上松範康

边境之国的半夏迷梦–花之祭P

ぽかぽかうさぎ日和–川田瑠夏

リンゴ日和~The Wolf Whistling Song–ROCKY CHACK

vits变声

https://github.com/w4123/vits

白嫖原神语音 有对如何训练原神语音作出的修改 想白嫖访问demo的api 这个模型读英语不行 可能需要原神英文语音包吧 当然也可以直接考虑变声器模型 读阿拉伯数字也不可以 如何把非汉字内容变为可以读的内容

阿拉伯数字转汉字funnlp看到的

原神语音包解压

米游社 原神语音包

obtain all genshin impact voices in all languages in quotes, with text

genshin voice scraper with text, also have scraped voice and text inside

genshin limited english voice

tactron2 变声 训练

MoeTTS是一个Tacotron2/HifiGAN模型+编译好的GUI版本发布仓库 有多个语音包

项目地址:https://github.com/luoyily/MoeTTS

主要的vst变声插件:

变声插件以及其他VST的选择,大家可以参考上一篇专栏,本人现在用的是Avox Mutator调整声调,Little Alterboy调整共振峰的搭配:

gachikoe! core sign in with pixiv to download for free

VST机架 变声 设置教程

需要多个vst插件相互连接

可能只适合个别声音 如果是野生音源 需要预处理 再进行输入

本篇教程是上一篇教程的延续,咕咕了大半年的我也做了很多新的尝试,本篇文章包含的内容是:1. 介绍一个免费的更加适合直播用途的机架Cantabile(替代Reaper);2. 介绍目前咱自己调出来的,可以实现比较自然变声效果的VST链。关于如何使用OBS矫正变声器延迟导致的与模型嘴型、歌回时伴奏不同步的问题,我会在下一期中介绍(我真的不会再咕咕半年了)。

关于Voicemeeter与声卡配置的补充

请参照上一期,完成硬件设备以及Voicemeeter Banana的配置(重要!),Reaper的部分用下文的Cantabile代替~

本文中将会使用上一篇专栏中外置声卡的Asio驱动+第二张声卡(或者使用电脑自带的输出,比如一般电脑都有Realtek High Definition Audio)的方案。打开Voicemeeter的设置界面,在Output A1中点击ASIO设备的名称的话,会跳出声卡的配置界面:

点击红框处

声卡配置界面 - 采样率(Sample Rage)

声卡配置界面 - 缓存大小(Buffer Size)

采样率建议选择48kHz足够满足直播用途,太高的话会提高音频处理时的资源占用;如果在之后配置完变声器发现有爆音的现象,可以尝试提高Buffer Size(推荐512左右)。需要注意的是,Voicemeeter的ASIO方案虽然效率很高但可能是因为驱动的原因,并不适用于所有声卡,比如Focusrite的Scarlett系列。咱财力有限,只测试了Steinberg的UR系列声卡,不过在官方论坛上暂时也只看到了对Focusrite驱动支持问题的反馈,所以理论上大部分常用的声卡应该没有问题(大概

至此,前期准备完成!

Cantabile配置

下载地址:https://www.cantabilesoftware.com/download/

下载Stable Build并完成安装后,根据自己的操作系统选择运行64位/32位版本,选择Cantabile Lite版本(免费)。如果要求注册账户,按照步骤使用邮箱完成注册后,会在邮箱内收到注册码,使用注册码激活即可。

选择“Cantabile 3 Lite”

进入Cantabile后,在菜单中选择Tools→Options,在Audio Engine中选择ASIO - Voicemeeter Virtual ASIO:

Audio Engine

将Audio Ports如下图配置(先按照上一篇专栏配置好Voicemeeter哦):

Audio Ports

同时在下面的Plugin Options中选择Add→Browse可以添加自己安装VST插件的文件夹,然后等待扫描完成,这样才可以使用已经安装的插件:

Plugins Options

在Cantabile的主界面中黑色部分右键选择Insert Plugin可以插入插件,右键已经插入的插件选择Delete可以删除该插件。实现变声器需要的插件的连接方法也非常简单,从Main Microphone拖出一条线到第一个插件的Stereo / Mono in,再将插件的Stereo / Mono Out连接到下一个插件的Stereo / Mono in,一直到最后一个插件的Stereo / Mono Out连接到Main Speakers。这样连接的原理很简单,就是将一个插件处理完的音频继续交由下一个插件处理,打到叠加的效果。下图是一个示例:

插件连接示例

为了方便监听效果,记得按照上一篇教程Voicemeeter配置部分的最后所说,保持Voicemeeter开启,点亮Voicemeeter绿框中的A2与B2,将Main Microphone到Main Speaker的线路连通应该就能从Voicemeeter右上角A2你选择的设备中听到此时经过处理的声音了,如果未连通应该是不会听到任何声音的。

至此,Cantabile配置完成!

搭建变声器所需插件

首先,介绍一下在搭建变声器的过程中需要的插件,下载与购买地址在文章末尾。在保证效果的前提下咱已经尽量把插件的选择精简,比如Waves、Soundtoys、Fabfilter的插件有捆绑包而不需要一个个购买,文章里没有提到的插件大家也可以自己玩一玩。个人的配置方案会在后文贴出:

(一)变声器插件(当然是必选)

经过一大堆尝试(Elastique Pitch,KeroVee,Antares Throat,等等…),效果最让咱满意的是来自Soundtoys的Little Alter Boy(售价$99,很贵,某些地方可以搜到Soundtoys的插件合辑但希望大家可以支持正版)。可以用的替代品是插件版的Gachikoe,需要加入作者桜音さち老师的Fanbox才能下载,要求一个月1000日元,下载完了可以取消,同样尽量支持作者~

变声器插件调节的是人声的音调(Pitch)以及共振峰(Formant)。在很多变声器插件中,Pitch的范围是-12到+12。一个八度有12个半音,从降一个八度到升一个八度,就是-12到+12这个范围的含义。咱推荐把Pitch直接调到+12,因为升高一个八度的话在歌回时不必另外对歌曲进行降调操作。相应的Formant请通过监听自己的声音慢慢调节,一般在3-4左右。如果想要自然一些并且不介意声音低沉一些的话,低一些的Pitch也是可以的。如何在OBS中进行对声音延迟的校正,让变声处理后的声音与耳机里听到的伴奏同步输出咱会写在下一篇专栏里~

(二)均衡器 / EQ(强烈推荐)

简单来说,均衡器调整的是各个频段的响度。由于男生和女生发生结构导致的响度分布不同,在调整音调的之前之后,我们都需要用到均衡器,不然可能会导致中频过强等问题。本人用的EQ是Fabfilter Pro Q3。

(三) 降噪器 / De-noiser(强烈推荐)

降噪器自然是为了处理噪声。在通过变声器之前,一些噪声可能并不刺耳,甚至可能都不会被注意到,但是在经过了变声器处理后,一些中低频的噪声会随着音调提高而变得刺耳,比如风声和电流声。因此,在声音通过变声器之前咱通常会放一个降噪器。本人用的是Waves WNS 1。

(四)限制器 / Limiter(强烈推荐)

限制器是其实可以说是压缩器的一种,将超出声音阈值的部分完全削除。在输出之前加一个-1分贝左右的限制器可以防止爆音以及大音量造成的失真保护观众的耳膜。咱用的是Waves Vcomp压缩器附带的限制器。

(五)Roth-AIR(推荐)

Roth-AIR是一个免费的、增加声音“空气感”的插件。本质上来说,应该属于EQ,但是使用非常方便(其实只需要调节一个旋钮)并且效果很好因此单独列出,咱在EQ之后会继续用Roth-Air调节高频。

(六)齿音消除器 / De-esser(推荐)

所谓齿音,是在说话或者唱歌时,开口说“嘶”、“次”这些音节时的尖锐声音。这些声音同样也会被变声器放大而变得刺耳。解决方法的方法,一种是使用插件,一种时给麦克风加上防喷网,也可结合使用。咱用的是防喷网+Waves DeEsser。

(七)压缩器 / Compressor(可选)

压缩器是当声音强度超过一定阈值时,对超出部分按照一定的比例进行响度衰减,以此减小声音的动态范围,同时不同的压缩器也会给声音带来不同的音染。希望让自己的声音更加饱满或者和麦克风距离忽近忽远导致声音忽大忽小的小伙伴可以尝试一下这类插件。本人在VST链最后加了Waves Vcomp插件。

介绍完毕,可以正式开工了!

正式配置变声器

大致的VST链顺序:

输入→降噪器→均衡器1→变声器→齿音消除器→EQ2→RothAIR→压缩器→输出

Cantabile配置

降噪插件的参数可以在自己不说话的情况下,点击频谱图右边的Suggest,会自动适配参数:

WNS 1 参数设置

第一个均衡器的配置,个人方案,请根据自己情况调节:

均衡器参数设置

变声器的配置,个人方案,请根据自己情况调节:

变声器配置

齿音消除器,第二个均衡器,RothAIR以及最后的压缩器参数就不贴图了,各位自己根据自己的声音多做尝试吧~

变声器教程完结!撒花!

一些大家可能会问的问题:

Q:只要有变声器就好了吗?

A:并不是,想要达到最好的效果还是需要控制自己的说话方式以及呼吸,之后可能会出专门的专栏讨论这个问题。

Q:这个方案的缺点?

A:资源占用不低,且有较大的延迟,不适合直播时监听变声过后的效果,因此需要练习把握住自己的语气。至于直播时变声器延迟导致的与伴奏不同步,与模型嘴型不同步我会在下期讲解怎么解决(咕咕咕)。以及,正版的插件价格很贵土真好吃,但还是希望大家尽可能支持正版。至于值不值得,大家可以听一下效果演示再做决定~

Q:除了文中提到的软件,是否有更好的选择?

A:有。本人自用的机架是Gig Performer 3($149),价格贵但是更加稳定,对VST3插件的支持更好。以及,Waves插件之外,iZotope全家桶的效果也很好,但是因为巨大的延迟,比起直播更加适合后期处理,有兴趣和钱的话可以研究一下。

写在最后的话:

首先,咱只是一个爱好者,并不是什么后期调音man,甚至耳朵也很木,但我写在这里的是我自己目前为止研究出来的,毫无保留的最佳方案。当然,这个方案有着巨大的改进余地,希望它能抛砖引玉,各位有更好的方案希望能不吝分享。写这篇专栏的原因是在搜寻的过程中看到了不少效果尚不尽如人意的变声器,甚至有打着“免费变声器”、“主播同款”名号的视频,提供QQ群号,提供破解的机架下载,然后推荐声卡,最后再收费调音,被质疑调完音效果仍然不好就消失换个号继续,所谓的演示视频也很明显是由一男一女分开录音的。当然,这些也是个别现象,我还是遇见了很多前辈们有启发性的视频,但大家还是要对这类陷阱提高警惕~

最后,谢谢看到这里的你!祝早日成为美少女

Soundtoys官网:

https://www.soundtoys.com/

Gachikoe插件版Fanbox链接:

https://sakuranesachi.fanbox.cc/posts/498211

Roth-AIR官网:

https://www.danielrothmann.com/#downloads

Fabfilter官网:

https://www.fabfilter.com/

Waves插件官网:

https://www.waves.com/

xidada’s tts:

https://huggingface.co/spaces/X*i-J*i*nP*i*n*g/X*i-J*i*nP*i*n*g-TTS # remove asterisks!

哔哩哔哩上看到的免费变声器

下载地址:

https://yuanqiyinpin.github.io/

本机架为二次开源软件,优点是占用率滴,不吃电脑配置,上手简单,免费使用

SoundTouch

萝莉音 青年音

https://github.com/jrising/pysoundtouch

gan based voice changer:

https://github.com/yl4579/StarGANv2-VC

install crossover to run windows app on linux

compile crossover from source on macos(code avaliable from official website):

https://gist.github.com/Alex4386/4cce275760367e9f5e90e2553d655309

https://www.codeweavers.com/crossover/source

变声器一般是vst类型的

run vst on linux headlessly:

https://github.com/hq9000/cython-vst-loader

https://github.com/hq9000/py_headless_daw

linux vst wrapper/bridge:

https://github.com/osxmidi/LinVst

VST bridge for Windows vst on Linux

https://github.com/abique/vst-bridge

use vst 2.4 on macos with obs studio:

https://github.com/obsproject/obs-vst

pyvst vst wrapper for windows:

https://github.com/mbrucher/PyVST

python vst2 wrapper for windows:

https://pypi.org/project/neil-vst/

yabridge use windows vst3, vst2 plugins on linux using wine, with reaper:

https://github.com/robbert-vdh/yabridge

lyrebird voice changer for linux gtk3:

https://github.com/lyrebird-voice-changer/lyrebird

voice changer based on MHW Audio Modding Tool (not recommended):

https://github.com/ItsBurpee/MHWVoiceChanger

mozilla voice changer and visualizer based on web api:

https://github.com/mdn/voice-change-o-matic

Real time voice changer in python:

https://github.com/symphonly/figaro

Pyvoicechanger:

https://github.com/juancarlospaco/pyvoicechanger

change the Pitch of the voice:

https://github.com/wittymindstech/change-voice-pitch

change pitch in real time:

https://github.com/jmt329/PitchShifter

Read More

2022-05-28
Ai上色

Read More

2022-05-27
推荐系统 Gnn

字节跳动 抖音推荐算法 Wide&Deep


1、muricoca/crab

https://github.com/muricoca/crab

2、ibayer/fastFM

https://github.com/ibayer/fastFM

3、Mendeley/mrec

https://github.com/mendeley/mrec

4、MrChrisJohnson/logistic-mf

https://github.com/MrChrisJohnson/logistic-mf

5、jadianes/winerama-recommender-tutorial

https://github.com/jadianes/winerama-recommender-tutorial

6、ocelma/python-recsys

https://github.com/ocelma/python-recsys

7、benfred/implicit

https://github.com/benfred/implicit

8、lyst/lightfm

https://github.com/lyst/lightfm

9、python-recsys/crab

https://github.com/python-recsys/crab

10、NicolasHug/Surprise

https://github.com/NicolasHug/Surprise


linkedin gdmix simple and memory effective personalized ranking

datawhale fun-rec 推荐系统入门教程

datawhale rechub

image to text, text to image, clip as image/text embeddings

deep recommendation using tensorflow 1.15

image recommendation system

不同的人有不同喜好

不同的人和不同的人说话

不同的产品有不同的特征

不同的产品和不同的产品被一起推荐

人对产品的接受度

youzan has an ai platform called trexpark, offering chinese NLP and image models pretrained from e-commerce databases.

https://github.com/youzanai/trexpark

session based recommendation system:

https://github.com/CRIPAC-DIG/SR-GNN

decide the feedback embeddings:

https://huggingface.co/youzanai/bert-product-comment-chinese

conversational embeddings:

https://huggingface.co/youzanai/bert-customer-message-chinese

neo4j developer build a recommendation engine:

https://neo4j.com/developer/cypher/guide-build-a-recommendation-engine/

torch_geometric(PyG) documentation:

https://pytorch-geometric.readthedocs.io/en/latest/modules/nn.html#torch_geometric.nn.conv.GatedGraphConv

setup GCN using PyG:

https://zhuanlan.zhihu.com/p/400078504

tagspace text classification via hashtags:

https://paddlerec.readthedocs.io/en/latest/models/contentunderstanding/tagspace.html

gnn is based on basic data/label models and provide high-level reasoning and predictions.

neo4j graph academy practical usage:

https://graphacademy.neo4j.com/categories/

https://neo4j.com/graphacademy/training-iga-40/12-iga-40-ingredient-analysis/

video segments have different features and orders. predict missing links. predict categories semi-supervised or unsupervised.

video-image-text-music correlation and predict internal relationships, categories.

recommendation system:

paddlerec(multimodal), torchrec(cuda==11.3, build failed due to unable to find ATen from torch/include.)

https://neo4j.com/docs/graph-data-science/current/end-to-end-examples/fastrp-knn-example/

link prediction:

https://github.com/Orbifold/pyg-link-prediction/blob/main/run.py

how to use pyg for link prediction:

https://github.com/pyg-team/pytorch_geometric/issues/634

dgl, install from source, with link prediction:

https://docs.dgl.ai/tutorials/blitz/4_link_predict.html

https://github.com/dmlc/dgl

https://docs.dgl.ai/guide_cn/training-link.html#guide-cn-training-link-prediction

gnn intro:

https://cnvrg.io/graph-neural-networks/

gnn applications:

Node classification: The objective here is to predict the labels of nodes by considering the labels of their neighbors.

Link prediction: In this case, the goal is to predict the relationship between various entities in a graph. This can for example be applied in prediction connections for social networks.

Graph clustering: This involves dividing the nodes of a graph into clusters. The partitioning can be done based on edge weights or edge distances or by considering the graphs as objects and grouping similar objects together.

Graph classification: This entails classifying a graph into a category. This can be applied in social network analysis and categorizing documents in natural language processing. Other applications in NLP include text classification, extracting semantic relationships between texts, and sequence labeling.

Computer vision: In the computer vision world, GNNs can be used to generate regions of interest for object detection. They can also be used in image classification whereby a scene graph is generated. The scene generation model then identifies objects in the image and the semantic relationship between them. Other applications in this field include interaction detection and region classification.

Read More

2022-05-24
Qq 微信 信息提取 Bot搭建

qq聊天记录导出 qq消息导出

微信聊天记录导出

聊天记录渲染成图片 render chat record to picture

conclusion so far: people like to use vue to recreate popular interfaces, and you may grab some interface from it.

vue-wechat

🔥 基于Vue2.0高仿微信App的单页应用

vue-qq

一个长得像QQ的demo

vue qq 聊天界面组件库

1
2
npm install vue-mchat

vue 本项目是一个在线聊天系统,最大程度的还原了Mac客户端QQ。

vue-miniQQ————基于Vue2实现的仿手机QQ单页面应用

基于Vue2实现的单页面应用 qq界面模仿

demo大师 qq界面模仿 vuejs 要钱

demo大师 vue3 仿微信/qq界面 免费

render chat record to picture 微信聊天记录渲染成图片

html css渲染

仿QQ android的聊天界面

HTML5手机微信聊天界面代码

HTML5 WebSocket 仿微信界面的网页群聊演示Demo

用html5做的仿微信聊天界面

基于H5技术实现的在线聊天室APP

Simple chatbot exercise using only JavaScript, HTML, CSS

Multi-Room Chat Application

一个基于AngularJS、NodeJS、Express、socket.io搭建的在线聊天室。

facebook like chatroom

qq空间发美女图片把人家的脸要挡住 或者要把脸换了 或者直接使用live2d three.js 甚至3d的渲染模型来把脸给它挡住

somehow the wechat web uos protocol is usable again? check it out.

https://www.npmjs.com/package/wechaty-puppet-wechat

https://github.com/wechaty/puppet-wechat/pull/206

would it be a lot easier if we can send those article/video links to external (out of gfw) social media platforms in their native language? still censorship will be applied.

wechat frida hook on macos:

https://github.com/dounine/wechat-frida

WeChat PC Frida hook:

https://github.com/K265/frida-wechat-sticker

https://github.com/kingking888/frida_wechat_hook

qq号码注册规则

qq群最多可以添加500个群 1500个好友 其中群可加的数量 = max(0,500 - 已加入群数量 - 好友数量)

可以退出一些安静的群 不发红包的群 删除好友

屏蔽别人加我为好友 允许别人拉我进群 自动退出广告群 退出不活跃的群

群一天只能加两三个 或者手机上可以加十个

好友一天可以加三十几个

一个验证QQ群的Python代码

https://www.bilibili.com/read/mobile?id=10044756

frida inject mobile android qq and open qzone:

https://github.com/xhtechxposed/fridainjectqq

search https://qun.qq.com in search engines

可以考虑截图获取QQ群验证问题 或者手机测试 appium

if possible then just use frida/radare2 or some reverse engineering to automate the process.

radare2 -> rizin.re(radare2 fork) based, ida alike, with ghidra decompiler, reverse engineering tool:

https://cutter.re

如何获取进群验证问题?记得可以拦截PC端搜索QQ群接收的数据包获取验证问题 或许不行 总之可以获取到一些参数 查看是否包含验证问题 是不是允许任何人进群 也可以考虑拦截opqqq的通信 或者发送一些通用的加群验证信息 比如“加群学习” “小伙伴一起玩” 之类的 或者用ai模型根据群描述 群主题 生成

一个手机号码可以申请10个qq号,一个手机号绑定的QQ帐号名额上限为10个,但一天一个手机号只能成功注册两到三个

WeChat needs serious reverse engineering like frida.

https://github.com/cixingguangming55555/wechat-bot

有webapi的微信机器人 注入dll到pc

https://github.com/mrsanshui/WeChatPYAPI

可以加好友的python wechat pc hook

https://github.com/snlie/WeChat-Hook

易语言的wechat hook 功能非常全 搜索 加人 有教程链接 教学代码

https://github.com/TonyChen56/WeChatRobot

比较老的wechat逆向模块 wechatapis.dll半天获得不了 有教程链接

https://github.com/wechaty/puppet-xp

frida 驱动的wechat puppet 暂时没有加人 搜索人 在windows上运行

wechat reverse engineering tutorials:

https://github.com/hedada-hc/pc_wechat_hook

https://github.com/zmrbak/PcWeChatHooK

wechaty base framework:

https://github.com/Wechaty/python-wechaty/ (puppet support might be incomplete)

https://github.com/Wechaty/wechaty/

botoy opqbot api for python

https://botoy.opqbot.com/zh_CN/latest/action/

qq opqbot (for wechat it has rstbot) download and install (need gitter.im api token):

https://docs.opqbot.com/guide/manual.html#启动失败

opqbot needs to be reverse engineered or we won’t know what is going on inside.

unofficial opqbot wiki:

https://mcenjoy.cn/opqbotwiki/

wechat bot(non-free wechat puppets):

wechaty

quoted content are controversial and highly viral. must be filtered and classified before proceeding.

quotes are like comments.

Read More

2022-05-24
Ai训练集标注工具

text annotation tool:

https://github.com/doccano/doccano

sqlite 3 backend:

1
2
pip3 install doccano

video/image annotation tool, needs docker, with online demo:

https://github.com/openvinotoolkit/cvat

image labeling:

https://github.com/heartexlabs/labelImg

with audio video support

https://github.com/heartexlabs/label-studio

with audio transcription support

https://github.com/UniversalDataTool/universal-data-tool

image and audio

https://github.com/Cartucho/OpenLabeling

specialized for yolo bounding boxes

https://github.com/developer0hye/Yolo_Label

Read More

2022-05-13
Attractive Dynamic Plus Attractive Video

Some contents are viral to the users. Will add extra watches if combined with related video or essay.

May apply the same rule to other platforms. Must select those with largest views, or verified by trained grading models. Native language only, or we have to translate and verify/convey it into native form. Post it to QQ, other platforms in the form of pictures, links.

Read More

2022-05-11
Anime Smile Detection_ Segmentation

Anime smile detection/ segmentation

when an anime head is detected, cut it out and create dataset with labels. may augmented it with grayscale or edge detection.

segmentation using labeled data and train it on pretrained models. using anme head detection as double verification. no double heads.

ppse recognition may be applied without further training, or else.

我分析需要YOLO确定人物位置 CNN判断服装类型 人物性别 ocr识别字幕 音频分析识别语气 性别 音乐类型 再用seq2seq来把所有的输出概括成我的描述

或者看看有没有文字转关键词的模型

可以的话加上人物姿态估计 动漫人物的

关于光流算法:

熵就是梯度的标准差

一段范围的熵就是起始时间到末尾的熵的标准差

或者起始到末尾的梯度的标准差

Read More