最近打算把 Jarvis 的部分外接功能从百度 AI 平台转到腾讯 AI 平台，但是腾讯 AI 并没有提供 SDK，所以自己写了一个分享出来。
这里面与百度 AI 平台之间主要的区别是腾讯的平台有一个鉴权机制，在调用 API 时除了一些基本参数之外，还需要提供一个根据 API 所需所有参数字典算出来的一个 MD5 码。格式以官方的 DEMO 作为参考并做了一定的改动，可读性相比 DEMO 有一定提高。
项目地址：Tencent_aiplat_SDK
源码如下：（与最终版本可能会有差别，以 GitHub 上的项目内代码为主）
# _*_ coding:utf-8 _*_
import hashlib
import urllib.parse
import urllib.request
import urllib.error
import json
import time
import base64
import os
# 接口api
url_prefix = 'https://api.ai.qq.com/fcgi-bin/'
class AiPlat(object):
def __init__(self, app_id, app_key):
self.app_id = app_id
self.app_key = app_key
self.data = {}
def genSignString(self, parser):
uri_str = ''
for key in sorted(parser.keys()):
if key == 'app_key':
continue
uri_str += "%s=%s&" % (key, urllib.parse.quote(str(parser[key]), safe=''))
sign_str = uri_str + 'app_key=' + self.app_key
hash_md5 = hashlib.md5(sign_str.encode("latin1"))
return hash_md5.hexdigest().upper()
def invoke(self, params):
self.url_data = urllib.parse.urlencode(params).encode(encoding='utf-8')
req = urllib.request.Request(self.url, self.url_data)
try:
rsp = urllib.request.urlopen(req)
str_rsp = rsp.read()
dict_rsp = json.loads(str_rsp.decode('utf-8'))
return dict_rsp
except urllib.error.URLError as e:
dict_error = {}
if hasattr(e, "code"):
dict_error = {}
dict_error['ret'] = -1
dict_error['httpcode'] = e.code
dict_error['msg'] = "sdk http post err"
return dict_error
if hasattr(e, "reason"):
dict_error['msg'] = 'sdk http post err'
dict_error['httpcode'] = -1
dict_error['ret'] = -1
return dict_error
# 智能闲聊
def getNlpTextChat(self, question):
self.data = {}
self.url = url_prefix + 'nlp/nlp_textchat'
self.data = {
'app_id': self.app_id,
'app_key': self.app_key,
'time_stamp': int(time.time()),
'nonce_str': int(time.time()),
'session': 10000,
'question': question
}
self.data['sign'] = self.genSignString(self.data)
return self.invoke(self.data)["data"]["answer"]
# 语音识别
def aai_asr(self, filepath): # filepath为音频文件的存档位置
res = str(base64.b64encode(open(filepath, "rb").read()), 'utf-8')
self.url = 'https://api.ai.qq.com/fcgi-bin/aai/aai_asr'
self.data = {
'app_id': self.app_id,
'app_key': self.app_key,
'time_stamp': int(time.time()),
'nonce_str': int(time.time()),
'format': 2,
'speech': res,
}
self.data['sign'] = self.genSignString(self.data)
return self.invoke(self.data)['data']['text']
# 语音合成
def aai_tts(self, sentence, filepath): # filepath为音频文件的存档位置
self.url = url_prefix + 'aai/aai_tts'
self.data = {
'app_id': self.app_id,
'app_key': self.app_key,
'time_stamp': int(time.time()),
'nonce_str': int(time.time()),
'speaker': 6,
'format': 2,
'volume': 0,
'speed': 100,
'text': sentence,
'aht': 0,
'apc': 58,
}
self.data['sign'] = self.genSignString(self.data)
result = self.invoke(self.data)['data']['speech']
with open(filepath, 'wb') as f:
res = base64.b64decode(result)
f.write(res)
f.close()
# 文字翻译
def textTrans(self, sentence):
self.url = url_prefix + 'nlp/nlp_texttrans'
self.data = {
'app_id': self.app_id,
'app_key': self.app_key,
'time_stamp': int(time.time()),
'nonce_str': int(time.time()),
'type': 0, # 设置翻译模式，英译中、中译英、英汉互译
'text': sentence,
}
self.data['sign'] = self.genSignString(self.data)
return self.invoke(self.data)['data']['trans_text']
# 语音翻译
def speechTrans(self, filepath): # filepath为音频文件的存档位置
res = str(base64.b64encode(open(filepath, "rb").read()), 'utf-8')
self.url = url_prefix + 'nlp/nlp_speechtranslate'
self.data = {
'app_id': self.app_id,
'app_key': self.app_key,
'time_stamp': int(time.time()),
'nonce_str': int(time.time()),
'format': 8, # MP3
'seq': 0,
'end': 1, # Finished or in line
'seq': 0,
'session_id': 'test1', # Name of the sppech sequence if it occur
'speech_chunk': res,
'source': 'zh', # source language
'target': 'en', # target language
}
self.data['sign'] = self.genSignString(self.data)
return [self.invoke(self.data)['data']['source_text'], self.invoke(self.data)['data']['target_text']]
def getAaiWxAsrs(self, filepath, speech_id, end_flag):
self.url = url_prefix + 'aai/aai_wxasrs'
file_len = os.path.getsize(filepath)
res = str(base64.b64encode(open(filepath, "rb").read()), 'utf-8')
self.data = {
'app_id': self.app_id,
'app_key': self.app_key,
'time_stamp': int(time.time()),
'nonce_str': int(time.time()),
'format': 2,
'rate': 16000,
'bits': 16,
'seq': 0,
'len': file_len,
'end': end_flag,
'speech_id': speech_id,
'speech_chunk': res,
'cont_res': 1,
}
self.data['sign'] = self.genSignString(self.data)
return self.invoke(self.data)['data']['speech_text']
client = AiPlat('', '')
print(client.textTrans('Hello'))
水平有限，若有误欢迎 Pull request，我会及时查阅并作改动。
