Implementation of word-marking translation browser script based on ChatGPT API

Preface

Recently there is a browser script based on ChatGPT API on GitHub, openai-translator. In a short period of time, the star has reached 12k. In addition to supporting translation, it also supports polishing and summarizing functions. In addition to In addition to the browser plug-in, tauri is also used to package a desktop client. Putting aside tauri and using the rust part, the browser part is relatively simple to implement. Today we will implement it manually.

Interface provided by openAI

For example, we can copy the following code and initiate a request in the browser console to complete the translation

//示例
const OPENAI_API_KEY = "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";

const prompt = `Translate this into Chinese:
hello world`;
const res = await fetch("https://api.openai.com/v1/completions", {
method: "POST",
headers: {
"Content-Type": "application/json",
authorization: `Bearer ${OPENAI_API_KEY}`,
},
body: JSON.stringify({
model: "text-davinci-003",
prompt,
max_tokens: 1000,
temperature: 0,
}),
});
const response = await res.json();

const result = response.choices[0].text;

OPONAI_API_KEY in the above code needs to be replaced with yours my own.

Realize word translation

Word translation is a common web page function. When the user selects a word or a text, a small window will automatically pop up to display the translation of the word or text.

1. First, add an empty DIV element and a button that triggers translation in the HTML page

let keyword;
const translation = document.createElement("div"); 
translation.id ="translation"; 
const icon = document.createElement("img"); 
icon.style.width ="30px";
icon.style.height = "30px";
icon.src ="http://example.com/icon.png";
translation.appendChild(icon)

2. Add a mouse-up event listener to the page , when the user selects a piece of text, set the search keywords.

document.addEventListener("mouseup", (event) => {
const selection = window.getSelection().toString().trim();
if (selection) {
keyword=selection；
}
});

3. Mouse click to execute translation logic. You can use AJAX requests to get the translation results from the background and display them in a DIV element.

function translate(){
if(keyword){
// 执行翻译逻辑
}
}
icon.addEventListener("mouseover", translate);

4. Add styles to the DIV element in the CSS style sheet so that it floats and displays on the page.

#translation {
position: fixed;
top: 10px;
right: 10px;
max-width: 300px;
padding: 5px;
background-color: #f7f7f7;
border: 1px solid #ccc;
box-shadow: 0px 2px 5px rgba(0, 0, 0, 0.1);
z-index: 9999;
}

The above steps can realize the basic function of word translation. Let’s take a look at the effect.

react antd implementation

The above code only implements the simplest version, and the style is also It is not beautiful enough, so we can use webpack react antd to implement a modern plug-in. Here I use a previously created template tampermonkey-starter.

Use antd's Popover component to display and use react to reconstruct the js code, we can achieve the following effects.

Word translation

Click the translation button, and the translation result will be displayed through the interface request below. However, the translation results will not be displayed until the api is fully returned, which will be slower. Can we use Stream? Does the OpenAI interface support stream rendering? In this way, the results will pop up word by word.

import { createParser } from "eventsource-parser";

const translate = async (text: string) => {
const resp = await fetch("https://api.openai.com/v1/completions", {
method: "POST",
headers: {
"Content-Type": "application/json",
authorization:
`Bearer ${OPENAI_API_KEY}`,
},
body: JSON.stringify({
model: "text-davinci-003",
prompt: `Translate this into Chinese:
${text}`,
max_tokens: 1000,
temperature: 0,
frequency_penalty: 0,
stream: true,
}),
});
if (resp.status !== 200) {
const res = await resp.json();
setLoading(false);
console.error(res);
return;
}
const parser = createParser((event) => {
if (event.type === "event") {
const data = event.data;
if (data === "[DONE]") {
setLoading(false);
}
try {
let json = JSON.parse(event.data);
setResult((prev) => {
return prev + json.choices[0].text;
});
} catch (error) {
console.log(error);
}
}
});
const data = resp.body;
if (!data) {
console.log("Error: No data received from API");
return;
}
const reader = resp.body.getReader();
try {
while (true) {
const { done, value } = await reader.read();
if (done) {
setLoading(false);
break;
}
const str = new TextDecoder().decode(value);
parser.feed(str);
}
} finally {
reader.releaseLock();
}
};

In the above code, we use the fetch API to send an HTTP request and obtain a readable stream in the response. We can use the getReader method to obtain a reader object and use it to process stream data, using the eventsource-parser package to parse data from server-sent events.

The content of the response will be displayed one by one according to Server-sent events (events sent by the server).

function speak(text) {
if ('speechSynthesis' in window) {
const utterance = new SpeechSynthesisUtterance(text);
utterance.voice = speechSynthesis.getVoices()[0];
utterance.pitch = 1;
utterance.rate = 1;
speechSynthesis.speak(utterance);
}
}

speak('Hello, world!');

The above is the detailed content of Implementation of word-marking translation browser script based on ChatGPT API. For more information, please follow other related articles on the PHP Chinese website!