ChatGPTSpeaker

A smart speaker that combines ChatGPT, Speech-to-Text, and Text-to-Speech.

How to use

create a config file config.ini

[porcupine]
access_key=XXXXXXinput_your_key_here

[openai]
api_key=XXXXXXinput_your_key_here

• visit https://picovoice.ai/ for porcupine key
• visit https://beta.openai.com/account/api-keys for openai key

install requirements

Mac

brew install portaudio
brew install mpg321
pip install -r requirements.txt

Linux

• The custom wake-up word model for this platform needs to be retrained, and the code also needs to load different wake-up word resources according to the platform selection. (It is probably unnecessary to use the default wake-up word).

Windows

Same as the Linux above.

run main.py

have fun

• the wake up word is "Hello Marry",so you should say "Hello Marry"
• ask a question
• you will get a response

enjoy it.

Some shortcomings to be improved

• TTS services require scientific networking for access which leads to instability.
• The overall time consumption needs to be optimized.
• The mute detection logic is not intelligent enough.
  • At the beginning: if speech is not given promptly when recording starts, recording may end directly.
  • At the end: if speech is given slowly during recording, it may be misjudged as ending.
• The code is not robust enough.
  • It does not support scenarios where the TTS network access is abnormal.
  • No try-catch is used in the overall code.
  • It feels a bit low to use a while true outside.
• we can only chat in Chinese now and haven't supported English conversation yet.

补充说明

这个工具能否在大陆使用呢？ 浏览器访问ChatGPT是不能在中国使用的，连香港也不行。 但是我们在科学上网后可以注册openAI的key，目前为止通过这个key是可以在国内访问的ChatGPT服务的。 但尴尬的事，我的stt 服务都是用的谷歌的服务，这个不行。在想办法代替。之前试过微软的 azure 未果。这个也算是todo 之一。

服务	是否离线	大陆能否访问	备注
唤醒词	离线
静音监测	离线
浏览器访问ChatGPT	网络	不能
通过key访问beta模型	网络	可以
语音转文字	网络	不能	目前用的谷歌，打算替换一个质量好且大陆可以访问的
文字转语音	网络	可以	目前用一个机器音，效果没有谷歌的好，效果有点类似mac 自带的say