谷歌人机交互页面：
https://www.recaptcha.net/recaptcha/api2/demo





 
如果直接从图片肝，需要收集足够的图片，然后使用yolo或者pytorch进行训练，得到模型后再进行识别。
 
不过这个人机交互验证码有一个语音验证的功能。
 
只要点击一个耳机的图标，然后就变成了语音识别。
播放一段录音，然后输入几个单词，如果单词对了，那么也可以通过。
 





 
那接下来的问题就简单了，拿到录音->识别录音，转化为文本，然后在输入框输入，就基本大功告成了。
 
英文转文本，网上有不少的AI平台可以白嫖，不过论效果，个人觉得百度的AI效果还不错，起码可以免费调用5W次。
 
完整代码如下：
 
这个是百度识别语音部分：
# -*- coding: utf-8 -*-
# @Time : 2021/4/24 20:50
# @File : baidu_voice_service.py
# @Author : Rocky C@www.30daydo.com

import os
import time

import requests
import sys
import pickle
sys.path.append('..')
from config import API_KEY,SECRET_KEY
from base64 import b64encode
from pathlib import PurePath
import subprocess

BASE = PurePath(__file__).parent

# 需要识别的文件
# 文件格式
# 文件后缀只支持 pcm/wav/amr 格式，极速版额外支持m4a 格式

CUID = '24057753' # 随意
# 采样率
RATE = 16000 # 固定值


ASR_URL = 'http://vop.baidu.com/server_api'

#测试自训练平台需要打开以下信息， 自训练平台模型上线后，您会看见 第二步：“”获取专属模型参数pid:8001，modelid:1234”，按照这个信息获取 dev_pid=8001，lm_id=1234
'''
http://vop.baidu.com/server_api
1537 普通话(纯中文识别) 输入法模型 有标点 支持自定义词库
1737 英语 英语模型 无标点 不支持自定义词库
1637 粤语 粤语模型 有标点 不支持自定义词库
1837 四川话 四川话模型 有标点 不支持自定义词库
1936 普通话远场
'''
DEV_PID = 1737

SCOPE = 'brain_enhanced_asr' # 有此scope表示有asr能力，没有请在网页里开通极速版


class DemoError(Exception):
pass


TOKEN_URL = 'http://openapi.baidu.com/oauth/2.0/token'

def fetch_token():

params = {'grant_type': 'client_credentials',
'client_id': API_KEY,
'client_secret': SECRET_KEY}
r = requests.post(
url=TOKEN_URL,
data=params
)

result = r.json()
if ('access_token' in result.keys() and 'scope' in result.keys()):
if SCOPE and (not SCOPE in result['scope'].split(' ')): # SCOPE = False 忽略检查
raise DemoError('scope is not correct')

return result['access_token']

else:
raise DemoError('MAYBE API_KEY or SECRET_KEY not correct: access_token or scope not found in token response')


""" TOKEN end """

def dump_token(token):
with open(os.path.join(BASE,'token.pkl'),'wb') as fp:
pickle.dump({'token':token},fp)

def load_token(filename):

if not os.path.exists(filename):
token=fetch_token()
dump_token(token)
return token
else:
with open(filename,'rb') as fp:
token = pickle.load(fp)
return token['token']

def recognize_service(token,filename):
FORMAT = filename[-3:]
with open(filename, 'rb') as speech_file:
speech_data = speech_file.read()

length = len(speech_data)
if length == 0:
raise DemoError('file %s length read 0 bytes' % filename)

b64_data = b64encode(speech_data)
params = {'cuid': CUID, 'token': token, 'dev_pid': DEV_PID,'speech':b64_data,'len':length,'format':FORMAT,'rate':RATE,'channel':1}

headers = {
'Content-Type':'application/json',
}
r = requests.post(url=ASR_URL,json=params,headers=headers)
return r.json()

def rate_convertor(filename):
filename = filename.split('.')[0]
CMD=f'ffmpeg.exe -y -i {filename}.mp3 -ac 1 -ar 16000 {filename}.wav'
try:
p=subprocess.Popen(CMD, stdin=subprocess.PIPE)
p.communicate()
time.sleep(1)
except Exception as e:
print(e)
return False,None
else:
return True,f'{filename}.wav'

def clear(file):
try:
os.remove(file)
except Exception as e:
print(e)

def get_voice_text(audio_file):

filename = 'token.pkl'
token = load_token(filename)
convert_status,file = rate_convertor(audio_file)
clear(file)
if not convert_status:
return None

result = recognize_service(token,file)

return result['result'][0]

if __name__ == '__main__':
get_voice_text('1.mp3')



 
然后下面的是获取语音部分，并且点击输入结果。
# -*- coding: utf-8 -*-
# @Time : 2021/4/25 15:16
# @File : download_mp3.py
# @Author : Rocky C@www.30daydo.com



#!/usr/bin/env python3
import os
import subprocess
import time
import re
import requests
import urllib.request
import zipfile
import io
from google.cloud import speech_v1
from random import randint, uniform
from bs4 import BeautifulSoup
from selenium import webdriver
from selenium.common.exceptions import NoSuchElementException
from selenium.webdriver.common.by import By
from selenium.webdriver.common.action_chains import ActionChains
from selenium.webdriver.support import expected_conditions as EC
from selenium.webdriver.support.ui import WebDriverWait
from google_recaptcha.baidu_voice_service import get_voice_text,clear


class Gcaptcha:

def __init__(self, url):
self.response = None

self.attempts = 0
self.successful = 0
self.failed = 0
self.solved = 0

self.mp3 = 'audio.mp3'
# self.wav='audio.wav'

# Set Chrome to run in headless mode and mute the audio
opts = webdriver.ChromeOptions()
# opts.headless = True
opts.add_argument("--mute-audio")
# opts.add_argument("--headless",)


CHROME_PATH = r'C:\git\EZProject\bin\chromedriver.exe'
self.driver = webdriver.Chrome(executable_path=CHROME_PATH,options=opts)
self.driver.maximize_window()
self.driver.get(url)
self.__bypass_webdriver_check()

# Initialize gcaptcha solver
self.__initialize()

while True:
# Download MP3 file
mp3_file = self.__download_mp3()

# Transcribe MP3 file
result = get_voice_text(self.mp3)

# audio_transcription = transcribe(mp3_file)
# self.transcription.attempts += 1

# If the MP3 file is properly transcribed
if result is not None:
# self.transcription.successful += 1

# Verify transcription
verify = self.__submit_transcription(result)

# Transcription successful with confidence >60%
if verify:
gcaptcha_response = self.__get_response()
self.response = gcaptcha_response
# self.recaptcha.solved += 1

# Delete MP3 file

self.driver.close()
self.driver.quit()
break
# Multiple correct solutions required. Solving again.
else:
self.solved += 1
clear(self.mp3)

# If the MP3 file could not be transcribed
else:
self.failed += 1
clear(self.mp3)

# Click on the "Get a new challenge" button to use a new MP3 file
self.__refresh_mp3()
# time.sleep(uniform(2, 4))



def __initialize(self):
# Access initial gcaptcha iframe
self.driver.switch_to.frame(self.driver.find_element(By.CSS_SELECTOR, 'iframe[name^=a]'))
self.__bypass_webdriver_check()

# Click the gcaptcha checkbox
checkbox = self.driver.find_element(By.CSS_SELECTOR, '#recaptcha-anchor')
self.__mouse_click(checkbox)

# Go back to original content to access second gcaptcha iframe
self.driver.switch_to.default_content()

# Wait roughly 3 seconds for second gcaptcha iframe to load
time.sleep(uniform(2.5, 3))

# Find second gcaptcha iframe
gcaptcha = WebDriverWait(self.driver, 10).until(
EC.presence_of_element_located((By.CSS_SELECTOR, 'iframe[name^=c]'))
)

# Access second gcaptcha iframe
self.driver.switch_to.frame(gcaptcha)
self.__bypass_webdriver_check()

# Click the audio button
audio_button = WebDriverWait(self.driver, 10).until(
EC.presence_of_element_located((By.CSS_SELECTOR, '.rc-button-audio'))
)
self.__mouse_click(audio_button)
time.sleep(0.5)

def __mouse_click(self, element):
cursor = ActionChains(self.driver)
cursor.move_to_element(element)
cursor.pause(uniform(0.3, 0.5))
cursor.click()
cursor.perform()

def __bypass_webdriver_check(self):
self.driver.execute_script(
'const newProto = navigator.__proto__; delete newProto.webdriver; navigator.__proto__ = newProto;')

def __download_mp3(self):
self.driver.switch_to.default_content()
self.driver.switch_to.frame(self.driver.find_element(By.CSS_SELECTOR, 'iframe[name^=c]'))
self.__bypass_webdriver_check()

# Check if the Google servers are blocking us
if len(self.driver.find_elements(By.CSS_SELECTOR, '.rc-doscaptcha-body-text')) == 0:
audio_file = WebDriverWait(self.driver, 10).until(
EC.presence_of_element_located((By.CSS_SELECTOR, '.rc-audiochallenge-tdownload-link'))
)

# Click the play button
play_button = WebDriverWait(self.driver, 10).until(
EC.presence_of_element_located((By.CSS_SELECTOR, '.rc-audiochallenge-play-button > button'))
)
self.__mouse_click(play_button)

# Get URL of MP3 file
audio_url = audio_file.get_attribute('href')

# Predefine the MP3 file name

# Download the MP3 file
try:
urllib.request.urlretrieve(audio_url, self.mp3)
except Exception as e:
print(e)
return None
else:
return self.mp3
else:
Error('Too many requests have been sent to Google. You are currently being blocked by their servers.')
exit(-1)

def __refresh_mp3(self):
self.driver.switch_to.default_content()
self.driver.switch_to.frame(self.driver.find_element(By.CSS_SELECTOR, 'iframe[name^=c]'))
self.__bypass_webdriver_check()

# Click on the refresh button to retrieve a new mp3 file
refresh_button = self.driver.find_element(By.CSS_SELECTOR, '#recaptcha-reload-button')
self.__mouse_click(refresh_button)

def __submit_transcription(self, text):
self.driver.switch_to.default_content()
self.driver.switch_to.frame(self.driver.find_element(By.CSS_SELECTOR, 'iframe[name^=c]'))
self.__bypass_webdriver_check()

# Input field for response
input_field = self.driver.find_element(By.CSS_SELECTOR, '#audio-response')

# Instantly type the full text without delays because Google isn't checking delays between keystrokes
input_field.send_keys(text)

# Click "Verify" button
verify_button = self.driver.find_element(By.CSS_SELECTOR, '#recaptcha-verify-button')
self.__mouse_click(verify_button)

# Wait roughly 3 seconds for verification to complete
time.sleep(uniform(2, 3))

self.driver.switch_to.default_content()
self.driver.switch_to.frame(self.driver.find_element(By.CSS_SELECTOR, 'iframe[name^=a]'))
self.__bypass_webdriver_check()

# Check to see if verified by recaptcha
try:
self.driver.find_element(By.CSS_SELECTOR, '.recaptcha-checkbox-checked')
except NoSuchElementException:
return False
else:
return True

def __get_response(self):
# Switch back to main parent window and get gcaptcha response
self.driver.switch_to.default_content()
response = self.driver.find_element(By.CSS_SELECTOR, '#g-recaptcha-response').get_attribute('value')
return response


class Error(Exception):
def __init__(self, message):
get_files = os.listdir()
match_regex = re.compile(r'^audio\d+.mp3|chromedriver_\w+\d+.zip$')
filtered_files = [f for f in get_files if match_regex.match(f)]

for file in filtered_files:
clear(file)

raise Exception(f'ERROR: {message}')


if __name__=='__main__':
gcaptcha = Gcaptcha('https://www.google.com/recaptcha/api2/demo')代码里需要你申请一个百度AI的key以便生成token。
最终试了，效果还是达到98%的准确率。 查看全部

# -*- coding: utf-8 -*-

# @Time : 2021/4/24 20:50

# @File : baidu_voice_service.py

# @Author : Rocky C@www.30daydo.com



import os

import time



import requests

import sys

import pickle

sys.path.append('..')

from config import API_KEY,SECRET_KEY

from base64 import b64encode

from pathlib import PurePath

import subprocess



BASE = PurePath(__file__).parent



# 需要识别的文件

# 文件格式

# 文件后缀只支持 pcm/wav/amr 格式，极速版额外支持m4a 格式



CUID = '24057753' # 随意

# 采样率

RATE = 16000  # 固定值





ASR_URL = 'http://vop.baidu.com/server_api'



#测试自训练平台需要打开以下信息， 自训练平台模型上线后，您会看见 第二步：“”获取专属模型参数pid:8001，modelid:1234”，按照这个信息获取 dev_pid=8001，lm_id=1234

'''

http://vop.baidu.com/server_api

1537	普通话(纯中文识别)	输入法模型	有标点	支持自定义词库

1737	英语	英语模型	无标点	不支持自定义词库

1637	粤语	粤语模型	有标点	不支持自定义词库

1837	四川话	四川话模型	有标点	不支持自定义词库

1936	普通话远场

'''

DEV_PID = 1737



SCOPE = 'brain_enhanced_asr'  # 有此scope表示有asr能力，没有请在网页里开通极速版





class DemoError(Exception):

    pass





TOKEN_URL = 'http://openapi.baidu.com/oauth/2.0/token'



def fetch_token():



    params = {'grant_type': 'client_credentials',

              'client_id': API_KEY,

              'client_secret': SECRET_KEY}

    r = requests.post(

        url=TOKEN_URL,

        data=params

    )



    result = r.json()

    if ('access_token' in result.keys() and 'scope' in result.keys()):

        if SCOPE and (not SCOPE in result['scope'].split(' ')):  # SCOPE = False 忽略检查

            raise DemoError('scope is not correct')



        return result['access_token']



    else:

        raise DemoError('MAYBE API_KEY or SECRET_KEY not correct: access_token or scope not found in token response')





"""  TOKEN end """



def dump_token(token):

    with open(os.path.join(BASE,'token.pkl'),'wb') as fp:

        pickle.dump({'token':token},fp)



def load_token(filename):



    if not os.path.exists(filename):

        token=fetch_token()

        dump_token(token)

        return token

    else:

        with open(filename,'rb') as fp:

            token = pickle.load(fp)

            return token['token']



def recognize_service(token,filename):

    FORMAT = filename[-3:]

    with open(filename, 'rb') as speech_file:

        speech_data = speech_file.read()



    length = len(speech_data)

    if length == 0:

        raise DemoError('file %s length read 0 bytes' % filename)



    b64_data = b64encode(speech_data)

    params = {'cuid': CUID, 'token': token, 'dev_pid': DEV_PID,'speech':b64_data,'len':length,'format':FORMAT,'rate':RATE,'channel':1}



    headers = {

        'Content-Type':'application/json',

    }

    r = requests.post(url=ASR_URL,json=params,headers=headers)

    return r.json()



def rate_convertor(filename):

    filename = filename.split('.')[0]

    CMD=f'ffmpeg.exe -y -i {filename}.mp3 -ac 1 -ar 16000 {filename}.wav'

    try:

        p=subprocess.Popen(CMD, stdin=subprocess.PIPE)

        p.communicate()

        time.sleep(1)

    except Exception as e:

        print(e)

        return False,None

    else:

        return True,f'{filename}.wav'



def clear(file):

    try:

        os.remove(file)

    except Exception as e:

        print(e)



def get_voice_text(audio_file):



    filename = 'token.pkl'

    token = load_token(filename)

    convert_status,file = rate_convertor(audio_file)

    clear(file)

    if not convert_status:

        return None



    result = recognize_service(token,file)



    return result['result'][0]



if __name__ == '__main__':

    get_voice_text('1.mp3')

# -*- coding: utf-8 -*-

# @Time : 2021/4/25 15:16

# @File : download_mp3.py

# @Author : Rocky C@www.30daydo.com







#!/usr/bin/env python3

import os

import subprocess

import time

import re

import requests

import urllib.request

import zipfile

import io

from google.cloud import speech_v1

from random import randint, uniform

from bs4 import BeautifulSoup

from selenium import webdriver

from selenium.common.exceptions import NoSuchElementException

from selenium.webdriver.common.by import By

from selenium.webdriver.common.action_chains import ActionChains

from selenium.webdriver.support import expected_conditions as EC

from selenium.webdriver.support.ui import WebDriverWait

from google_recaptcha.baidu_voice_service import get_voice_text,clear





class Gcaptcha:



    def __init__(self, url):

        self.response = None



        self.attempts = 0

        self.successful = 0

        self.failed = 0

        self.solved = 0



        self.mp3 = 'audio.mp3'

        # self.wav='audio.wav'



        # Set Chrome to run in headless mode and mute the audio

        opts = webdriver.ChromeOptions()

        # opts.headless = True

        opts.add_argument("--mute-audio")

        # opts.add_argument("--headless",)



      

        CHROME_PATH = r'C:\git\EZProject\bin\chromedriver.exe'

        self.driver = webdriver.Chrome(executable_path=CHROME_PATH,options=opts)

        self.driver.maximize_window()

        self.driver.get(url)

        self.__bypass_webdriver_check()



        # Initialize gcaptcha solver

        self.__initialize()



        while True:

            # Download MP3 file

            mp3_file = self.__download_mp3()



            # Transcribe MP3 file

            result = get_voice_text(self.mp3)



            # audio_transcription = transcribe(mp3_file)

            # self.transcription.attempts += 1



            # If the MP3 file is properly transcribed

            if result is not None:

                # self.transcription.successful += 1



                # Verify transcription

                verify = self.__submit_transcription(result)



                # Transcription successful with confidence >60%

                if verify:

                    gcaptcha_response = self.__get_response()

                    self.response = gcaptcha_response

                    # self.recaptcha.solved += 1



                    # Delete MP3 file



                    self.driver.close()

                    self.driver.quit()

                    break

                # Multiple correct solutions required. Solving again.

                else:

                    self.solved += 1

                    clear(self.mp3)



            # If the MP3 file could not be transcribed

            else:

                self.failed += 1

                clear(self.mp3)



                # Click on the "Get a new challenge" button to use a new MP3 file

                self.__refresh_mp3()

                # time.sleep(uniform(2, 4))







    def __initialize(self):

        # Access initial gcaptcha iframe

        self.driver.switch_to.frame(self.driver.find_element(By.CSS_SELECTOR, 'iframe[name^=a]'))

        self.__bypass_webdriver_check()



        # Click the gcaptcha checkbox

        checkbox = self.driver.find_element(By.CSS_SELECTOR, '#recaptcha-anchor')

        self.__mouse_click(checkbox)



        # Go back to original content to access second gcaptcha iframe

        self.driver.switch_to.default_content()



        # Wait roughly 3 seconds for second gcaptcha iframe to load

        time.sleep(uniform(2.5, 3))



        # Find second gcaptcha iframe

        gcaptcha = WebDriverWait(self.driver, 10).until(

            EC.presence_of_element_located((By.CSS_SELECTOR, 'iframe[name^=c]'))

        )



        # Access second gcaptcha iframe

        self.driver.switch_to.frame(gcaptcha)

        self.__bypass_webdriver_check()



        # Click the audio button

        audio_button = WebDriverWait(self.driver, 10).until(

            EC.presence_of_element_located((By.CSS_SELECTOR, '.rc-button-audio'))

        )

        self.__mouse_click(audio_button)

        time.sleep(0.5)



    def __mouse_click(self, element):

        cursor = ActionChains(self.driver)

        cursor.move_to_element(element)

        cursor.pause(uniform(0.3, 0.5))

        cursor.click()

        cursor.perform()



    def __bypass_webdriver_check(self):

        self.driver.execute_script(

            'const newProto = navigator.__proto__; delete newProto.webdriver; navigator.__proto__ = newProto;')



    def __download_mp3(self):

        self.driver.switch_to.default_content()

        self.driver.switch_to.frame(self.driver.find_element(By.CSS_SELECTOR, 'iframe[name^=c]'))

        self.__bypass_webdriver_check()



        # Check if the Google servers are blocking us

        if len(self.driver.find_elements(By.CSS_SELECTOR, '.rc-doscaptcha-body-text')) == 0:

            audio_file = WebDriverWait(self.driver, 10).until(

                EC.presence_of_element_located((By.CSS_SELECTOR, '.rc-audiochallenge-tdownload-link'))

            )



            # Click the play button

            play_button = WebDriverWait(self.driver, 10).until(

                EC.presence_of_element_located((By.CSS_SELECTOR, '.rc-audiochallenge-play-button > button'))

            )

            self.__mouse_click(play_button)



            # Get URL of MP3 file

            audio_url = audio_file.get_attribute('href')



            # Predefine the MP3 file name



            # Download the MP3 file

            try:

                urllib.request.urlretrieve(audio_url, self.mp3)

            except Exception as e:

                print(e)

                return None

            else:

                return self.mp3

        else:

            Error('Too many requests have been sent to Google. You are currently being blocked by their servers.')

            exit(-1)



    def __refresh_mp3(self):

        self.driver.switch_to.default_content()

        self.driver.switch_to.frame(self.driver.find_element(By.CSS_SELECTOR, 'iframe[name^=c]'))

        self.__bypass_webdriver_check()



        # Click on the refresh button to retrieve a new mp3 file

        refresh_button = self.driver.find_element(By.CSS_SELECTOR, '#recaptcha-reload-button')

        self.__mouse_click(refresh_button)



    def __submit_transcription(self, text):

        self.driver.switch_to.default_content()

        self.driver.switch_to.frame(self.driver.find_element(By.CSS_SELECTOR, 'iframe[name^=c]'))

        self.__bypass_webdriver_check()



        # Input field for response

        input_field = self.driver.find_element(By.CSS_SELECTOR, '#audio-response')



        # Instantly type the full text without delays because Google isn't checking delays between keystrokes

        input_field.send_keys(text)



        # Click "Verify" button

        verify_button = self.driver.find_element(By.CSS_SELECTOR, '#recaptcha-verify-button')

        self.__mouse_click(verify_button)



        # Wait roughly 3 seconds for verification to complete

        time.sleep(uniform(2, 3))



        self.driver.switch_to.default_content()

        self.driver.switch_to.frame(self.driver.find_element(By.CSS_SELECTOR, 'iframe[name^=a]'))

        self.__bypass_webdriver_check()



        # Check to see if verified by recaptcha

        try:

            self.driver.find_element(By.CSS_SELECTOR, '.recaptcha-checkbox-checked')

        except NoSuchElementException:

            return False

        else:

            return True



    def __get_response(self):

        # Switch back to main parent window and get gcaptcha response

        self.driver.switch_to.default_content()

        response = self.driver.find_element(By.CSS_SELECTOR, '#g-recaptcha-response').get_attribute('value')

        return response





class Error(Exception):

    def __init__(self, message):

        get_files = os.listdir()

        match_regex = re.compile(r'^audio\d+.mp3|chromedriver_\w+\d+.zip$')

        filtered_files = [f for f in get_files if match_regex.match(f)]



        for file in filtered_files:

            clear(file)



        raise Exception(f'ERROR: {message}')





if __name__=='__main__':

    gcaptcha = Gcaptcha('https://www.google.com/recaptcha/api2/demo')