优化旁白问题
All checks were successful
Gitea Actions Demo / Explore-Gitea-Actions (push) Successful in 2m54s
All checks were successful
Gitea Actions Demo / Explore-Gitea-Actions (push) Successful in 2m54s
This commit is contained in:
parent
27234f1a5a
commit
06aa52f152
@ -1,5 +1,35 @@
|
||||
// 以流式方式请求LLM大模型接口,并打印流式返回内容
|
||||
|
||||
// 过滤旁白内容的函数
|
||||
function filterNarration(text) {
|
||||
if (!text) return text;
|
||||
|
||||
// 匹配各种括号内的旁白内容
|
||||
// 包括:()、【】、[]、{}、〈〉、《》等
|
||||
const narrationPatterns = [
|
||||
/([^)]*)/g, // 中文圆括号
|
||||
/\([^)]*\)/g, // 英文圆括号
|
||||
/【[^】]*】/g, // 中文方括号
|
||||
/\[[^\]]*\]/g, // 英文方括号
|
||||
/\{[^}]*\}/g, // 花括号
|
||||
/〈[^〉]*〉/g, // 中文尖括号
|
||||
/《[^》]*》/g, // 中文书名号
|
||||
/<[^>]*>/g // 英文尖括号
|
||||
];
|
||||
|
||||
let filteredText = text;
|
||||
|
||||
// 逐个应用过滤规则
|
||||
narrationPatterns.forEach(pattern => {
|
||||
filteredText = filteredText.replace(pattern, '');
|
||||
});
|
||||
|
||||
// 清理多余的空格和换行
|
||||
filteredText = filteredText.replace(/\s+/g, ' ').trim();
|
||||
|
||||
return filteredText;
|
||||
}
|
||||
|
||||
async function requestLLMStream({ apiKey, model, messages, onSegment }) {
|
||||
const response = await fetch('https://ark.cn-beijing.volces.com/api/v3/bots/chat/completions', {
|
||||
method: 'POST',
|
||||
@ -54,7 +84,14 @@ async function requestLLMStream({ apiKey, model, messages, onSegment }) {
|
||||
// 处理最后的待处理文本(无论长度是否大于5个字)
|
||||
if (pendingText.trim() && onSegment) {
|
||||
console.log('处理最后的待处理文本:', pendingText.trim());
|
||||
await onSegment(pendingText.trim(), true);
|
||||
// 过滤旁白内容
|
||||
const filteredText = filterNarration(pendingText.trim());
|
||||
if (filteredText.trim()) {
|
||||
console.log('过滤旁白后的最后文本:', filteredText);
|
||||
await onSegment(filteredText, true);
|
||||
} else {
|
||||
console.log('最后的文本被完全过滤,跳过');
|
||||
}
|
||||
}
|
||||
continue;
|
||||
}
|
||||
@ -67,10 +104,13 @@ async function requestLLMStream({ apiKey, model, messages, onSegment }) {
|
||||
pendingText += deltaContent;
|
||||
console.log('LLM内容片段:', deltaContent);
|
||||
|
||||
// 检查是否包含分段分隔符
|
||||
if (segmentDelimiters.test(pendingText)) {
|
||||
// 按分隔符分割文本
|
||||
const segments = pendingText.split(segmentDelimiters);
|
||||
// 先过滤旁白,再检查分段分隔符
|
||||
const filteredPendingText = filterNarration(pendingText);
|
||||
|
||||
// 检查过滤后的文本是否包含分段分隔符
|
||||
if (segmentDelimiters.test(filteredPendingText)) {
|
||||
// 按分隔符分割已过滤的文本
|
||||
const segments = filteredPendingText.split(segmentDelimiters);
|
||||
|
||||
// 重新组合处理:只处理足够长的完整段落
|
||||
let accumulatedText = '';
|
||||
@ -81,7 +121,7 @@ async function requestLLMStream({ apiKey, model, messages, onSegment }) {
|
||||
if (segment) {
|
||||
accumulatedText += segment;
|
||||
// 找到分隔符
|
||||
const delimiterMatch = pendingText.match(segmentDelimiters);
|
||||
const delimiterMatch = filteredPendingText.match(segmentDelimiters);
|
||||
if (delimiterMatch) {
|
||||
accumulatedText += delimiterMatch[0];
|
||||
}
|
||||
@ -89,17 +129,22 @@ async function requestLLMStream({ apiKey, model, messages, onSegment }) {
|
||||
// 如果累积文本长度大于5个字,处理它
|
||||
if (accumulatedText.length > 8 && onSegment) {
|
||||
console.log('检测到完整段落:', accumulatedText);
|
||||
// 文本已经过滤过旁白,直接使用
|
||||
if (accumulatedText.trim()) {
|
||||
console.log('处理过滤后的文本:', accumulatedText);
|
||||
await onSegment(accumulatedText, false);
|
||||
}
|
||||
hasProcessed = true;
|
||||
accumulatedText = ''; // 重置
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 更新pendingText
|
||||
// 更新pendingText - 使用原始文本但需要相应调整
|
||||
if (hasProcessed) {
|
||||
// 保留未处理的累积文本和最后一个不完整段落
|
||||
pendingText = accumulatedText + (segments[segments.length - 1] || '');
|
||||
// 计算已处理的原始文本长度,更新pendingText
|
||||
const processedLength = pendingText.length - (segments[segments.length - 1] || '').length;
|
||||
pendingText = pendingText.substring(processedLength);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user