LangHelper
1.0.0
talk to ChatGPT / Speaking dialogue
support different speech types, web, AI with hundreds accents (VCTK corpus), will support azure, iflytek.
support two recognition tech, web and iflytek.
Taylor, Robert Downey, talk to any celebrations, such as Obama, Taylor, Downey...
will offer interface for loading your TTS model.
Speaking Assessment / Pronunciation Rating
integrate two Assessment API, speechsuper and iflytek now, open to get better algorithms to perfect the feature for IELTS/TOEFL test.
ChatGPT and LangHelper + espeak-ng Install x86 version. ChatGPT and LangHelper are both installed without installation, and espeak-ng can be installed by default. Video tutorial + text tutorial and precautions
Currently, to use AI pronunciation, you need to install ChatGPT desktop for Langhelper + Langhelper + espeak-ng. Because some libraries used to run AI are relatively large, I gave up packaging python programs into bin files and integrating them into ChatGPT desktop for Langhelper. espeak-ng is a must-reliance for speech synthesis.
Unzip the compressed package, a Langhelper folder (including langhelper.exe), and a chatgpt.exe, open respectively. langhelper will wait for chatgpt to complete before working. Chatgpt will set voice-related functions under -> preferences -> control center -> setting -> LangHelper. Without the configuration of the API, do not select other speech types and recognition. After the settings are completed, click submit, and there is a restart prompt to click yes. This is the langhelper window that will appear init -> start conversation, indicating that the initialization is completed.
The conversation cannot pronounce AI normally. First, click on the try to test whether the pronunciation can be performed. If the text is empty, the default synthesis will be "you are gorgeous, i love you". See what the langhlper interface prompts. error:Obama, means that the Obama accent cannot be used, and the other reason is that the environment may not be configured well. Please give detailed error screenshot information when sending questions in a group. The following is the synthesis How can I assist you today? Normal information prompts:
Text splitted to sentences. ['How can I assist you today?'] Hello! p241 Text splitted to sentences. ['Hello!'] Processing time: 0.8514664173126221 Real-time factor: 0.5273829916220033 Processing time: 0.6864285469055176 Real-time factor: 0.6285610240559246
For voice recognition, you need to click speech recognition first, and then click to end the recognition once.
How to set up prompts? Preferences -> control center -> Language Model -> User Custom -> Add model, add your custom prompts, here is my presetting ITELTS prompt for reference:
I want you to ask me some questions for simulating IELTS speaking test, non official but give score of reference,when you ask me a question, I'll respond you my spoken text and prounciation scores which come from speech recognition and assessment tech, please remember that you should ask me question one by one it means that you should offer another question after I give you response text which includes prouncation scores, do not give questions one time because I do not want to respond it by one time, after all questions finished , you can combine the score to assess my answer, please give me the final score of IELTS speaking test.so let's start first question.
11.22 OpenAI has released a completely free voice conversation function, which can be used in non-PLUS: https://twitter.com/OpenAI/status/1727065166188274145
