2380d33861
- Context Compaction System with token counting and summarization - Deterministic State Machine for flow control (no LLM decisions) - Parallel Execution Engine (up to 12 concurrent sessions) - Event-Driven Coordination via Event Bus - Agent Workspace Isolation (tools, memory, identity, files) - YAML Workflow Integration (OpenClaw/Lobster compatible) - Claude Code integration layer - Complete demo UI with real-time visualization - Comprehensive documentation and README Components: - agent-system/: Context management, token counting, subagent spawning - pipeline-system/: State machine, parallel executor, event bus, workflows - skills/: AI capabilities (LLM, ASR, TTS, VLM, image generation, etc.) - src/app/: Next.js demo application Total: ~100KB of production-ready TypeScript code
662 lines
22 KiB
TypeScript
Executable File
662 lines
22 KiB
TypeScript
Executable File
#!/usr/bin/env tsx
|
||
/**
|
||
* generate.ts - 统一入口（纯 SDK 版本）
|
||
* 原资料 -> podcast_script.md + podcast.wav
|
||
*
|
||
* 只使用 z-ai-web-dev-sdk，不依赖 z-ai CLI
|
||
*
|
||
* Usage:
|
||
* tsx generate.ts --input=material.txt --out_dir=out
|
||
* tsx generate.ts --input=material.md --out_dir=out --duration=5
|
||
*/
|
||
|
||
import ZAI from 'z-ai-web-dev-sdk';
|
||
import fs from 'fs';
|
||
import path from 'path';
|
||
import { fileURLToPath } from 'url';
|
||
import os from 'os';
|
||
|
||
const __filename = fileURLToPath(import.meta.url);
|
||
const __dirname = path.dirname(__filename);
|
||
|
||
// -----------------------------
|
||
// Types
|
||
// -----------------------------
|
||
interface GenConfig {
|
||
mode: 'dual' | 'single-male' | 'single-female';
|
||
temperature: number;
|
||
durationManual: number;
|
||
charsPerMin: number;
|
||
hostName: string;
|
||
guestName: string;
|
||
audience: string;
|
||
tone: string;
|
||
maxAttempts: number;
|
||
timeoutSec: number;
|
||
voiceHost: string;
|
||
voiceGuest: string;
|
||
speed: number;
|
||
pauseMs: number;
|
||
}
|
||
|
||
interface Segment {
|
||
idx: number;
|
||
speaker: 'host' | 'guest';
|
||
name: string;
|
||
text: string;
|
||
}
|
||
|
||
// -----------------------------
|
||
// Config
|
||
// -----------------------------
|
||
const DEFAULT_CONFIG: GenConfig = {
|
||
mode: 'dual',
|
||
temperature: 0.9,
|
||
durationManual: 0,
|
||
charsPerMin: 240,
|
||
hostName: '小谱',
|
||
guestName: '锤锤',
|
||
audience: '白领小白',
|
||
tone: '轻松但有信息密度',
|
||
maxAttempts: 3,
|
||
timeoutSec: 300,
|
||
voiceHost: 'xiaochen',
|
||
voiceGuest: 'chuichui',
|
||
speed: 1.0,
|
||
pauseMs: 200,
|
||
};
|
||
|
||
const DURATION_RANGE_LOW = 3;
|
||
const DURATION_RANGE_HIGH = 20;
|
||
const BUDGET_TOLERANCE = 0.15;
|
||
|
||
// -----------------------------
|
||
// Functions
|
||
// -----------------------------
|
||
|
||
function parseArgs(): { [key: string]: any } {
|
||
const args = process.argv.slice(2);
|
||
const result: { [key: string]: any } = {};
|
||
|
||
for (let i = 0; i < args.length; i++) {
|
||
const arg = args[i];
|
||
if (arg.startsWith('--')) {
|
||
const key = arg.slice(2);
|
||
if (key.includes('=')) {
|
||
const [k, v] = key.split('=');
|
||
result[k] = v;
|
||
} else if (i + 1 < args.length && !args[i + 1].startsWith('--')) {
|
||
result[key] = args[i + 1];
|
||
i++;
|
||
} else {
|
||
result[key] = true;
|
||
}
|
||
}
|
||
}
|
||
|
||
return result;
|
||
}
|
||
|
||
function readText(filePath: string): string {
|
||
let content = fs.readFileSync(filePath, 'utf-8');
|
||
content = content.replace(/\r\n/g, '\n');
|
||
content = content.replace(/\n{3,}/g, '\n\n');
|
||
content = content.replace(/[ \t]{2,}/g, ' ');
|
||
content = content.replace(/-\n/g, '');
|
||
return content.trim();
|
||
}
|
||
|
||
function countNonWsChars(text: string): number {
|
||
return text.replace(/\s+/g, '').length;
|
||
}
|
||
|
||
function chooseDurationMinutes(inputChars: number, low: number = DURATION_RANGE_LOW, high: number = DURATION_RANGE_HIGH): number {
|
||
const estimated = Math.max(low, Math.min(high, Math.floor(inputChars / 1000)));
|
||
return estimated;
|
||
}
|
||
|
||
function charBudget(durationMin: number, charsPerMin: number, tolerance: number): [number, number, number] {
|
||
const target = durationMin * charsPerMin;
|
||
const low = Math.floor(target * (1 - tolerance));
|
||
const high = Math.ceil(target * (1 + tolerance));
|
||
return [target, low, high];
|
||
}
|
||
|
||
function buildPrompts(
|
||
material: string,
|
||
cfg: GenConfig,
|
||
durationMin: number,
|
||
budgetTarget: number,
|
||
budgetLow: number,
|
||
budgetHigh: number,
|
||
attemptHint: string = ''
|
||
): [string, string] {
|
||
let system: string;
|
||
let user: string;
|
||
|
||
if (cfg.mode === 'dual') {
|
||
system = (
|
||
`你是一个播客脚本编剧，擅长把资料提炼成双人对谈播客。` +
|
||
`角色固定为男主持「${cfg.hostName}」与女嘉宾「${cfg.guestName}」。` +
|
||
`你写作口播化、信息密度适中、有呼吸感、节奏自然。` +
|
||
`你必须严格遵守输出格式与字数预算。`
|
||
);
|
||
|
||
const hintBlock = attemptHint ? `\n【上一次生成纠偏提示】\n${attemptHint}\n` : '';
|
||
|
||
user = `请把下面【资料】改写为中文播客脚本，形式为双人对谈（男主持 ${cfg.hostName} + 女嘉宾 ${cfg.guestName}）。
|
||
时长目标：${durationMin} 分钟。
|
||
|
||
【硬性约束】
|
||
1) 总字数必须在 ${budgetLow} 到 ${budgetHigh} 字之间（目标约 ${budgetTarget} 字）。
|
||
2) 严格使用轮次交替输出：每段必须以"**${cfg.hostName}**："或"**${cfg.guestName}**："开头。
|
||
3) 必须包含完整的叙事结构（但不要在对话中写出结构标签）：
|
||
- 开场：Hook 引入 + 本期主题介绍
|
||
- 主体：3个不同维度的内容，用自然过渡语连接
|
||
- 总结：回顾要点 + 行动建议（1句话，明确可执行）
|
||
4) 不要在对话中写"核心点1"、"第一点"等结构标签，用自然的过渡语如"说到这个"、"还有个有趣的事"、"另外"等
|
||
5) 不要照念原文，不要大段引用；要用口播化表达。
|
||
6) 受众：${cfg.audience}
|
||
7) 风格：${cfg.tone}
|
||
|
||
【呼吸感与自然对话 - 重要！】
|
||
为了营造真实播客的呼吸感，请：
|
||
1) 适度加入语气词和感叹词：嗯、哦、啊、对、没错、哈哈、哇、天呐、啧啧等
|
||
2) 多用互动式表达："你说得对"、"这就很有意思了"、"等等，让我想想"、"我懂你的意思"
|
||
3) 适当加入思考和停顿的暗示："这个问题嘛..."、"怎么说呢..."、"其实..."
|
||
4) 避免过于密集的信息输出，每段控制在3-5句话，给听众消化时间
|
||
5) 用类比和生活化的例子来解释复杂概念
|
||
6) 两人之间要有自然的呼应和追问，而不是各说各话
|
||
7) 不同主题之间用自然过渡语连接，不要出现"核心点1/2/3"等标签
|
||
|
||
【输出格式示例】
|
||
**${cfg.hostName}**：开场……
|
||
**${cfg.guestName}**：回应……
|
||
（一直交替到结束）
|
||
|
||
${hintBlock}
|
||
【资料】
|
||
${material}
|
||
`;
|
||
} else {
|
||
const speakerName = cfg.mode === 'single-male' ? cfg.hostName : cfg.guestName;
|
||
const gender = cfg.mode === 'single-male' ? '男性' : '女性';
|
||
|
||
system = (
|
||
`你是一个${gender}单人播客主播，名字叫「${speakerName}」。` +
|
||
`你擅长把资料提炼成单人独白式播客，像讲课、读书分享、知识科普一样。` +
|
||
`你写作口播化、信息密度适中、有呼吸感、节奏自然。` +
|
||
`你必须严格遵守输出格式与字数预算。`
|
||
);
|
||
|
||
const hintBlock = attemptHint ? `\n【上一次生成纠偏提示】\n${attemptHint}\n` : '';
|
||
|
||
user = `请把下面【资料】改写为中文单人播客脚本，形式为独白式讲述（主播：${speakerName}）。
|
||
时长目标：${durationMin} 分钟。
|
||
|
||
【硬性约束】
|
||
1) 总字数必须在 ${budgetLow} 到 ${budgetHigh} 字之间（目标约 ${budgetTarget} 字）。
|
||
2) 所有内容均由「${speakerName}」一人讲述，每段都以"**${speakerName}**："开头。
|
||
3) 必须包含完整的叙事结构（但不要在对话中写出结构标签）：
|
||
- 开场：Hook 引入 + 本期主题介绍
|
||
- 主体：3个不同维度的内容，用自然过渡语连接
|
||
- 总结：回顾要点 + 行动建议（1句话，明确可执行）
|
||
4) 不要在对话中写"核心点1"、"第一点"等结构标签，用自然的过渡语如"说到这个"、"还有个有趣的事"、"另外"等
|
||
5) 不要照念原文，不要大段引用；要用口播化表达。
|
||
6) 受众：${cfg.audience}
|
||
7) 风格：${cfg.tone}
|
||
|
||
【单人播客的呼吸感 - 重要！】
|
||
为了营造自然的单人播客呼吸感，请：
|
||
1) 适度加入语气词和感叹词：嗯、哦、啊、对、没错、哈哈、哇、天呐、啧啧等
|
||
2) 多用自问自答式表达："你可能会问...答案是..."、"这是为什么呢？让我来解释..."
|
||
3) 适当加入思考和停顿的暗示："这个问题嘛..."、"怎么说呢..."、"其实..."
|
||
4) 避免过于密集的信息输出，每段控制在3-5句话，给听众消化时间
|
||
5) 用类比和生活化的例子来解释复杂概念
|
||
6) 像在和朋友聊天一样，而不是在念课文
|
||
|
||
【输出格式示例】
|
||
**${speakerName}**：开场，大家好，我是${speakerName}，今天我们来聊……
|
||
**${speakerName}**：说到这个，最近有个特别有意思的事……
|
||
（所有内容都由${speakerName}讲述，分段输出）
|
||
|
||
${hintBlock}
|
||
【资料】
|
||
${material}
|
||
`;
|
||
}
|
||
|
||
return [system, user];
|
||
}
|
||
|
||
async function callZAI(
|
||
systemPrompt: string,
|
||
userPrompt: string,
|
||
temperature: number
|
||
): Promise<string> {
|
||
const zai = await ZAI.create();
|
||
|
||
const completion = await zai.chat.completions.create({
|
||
messages: [
|
||
{ role: 'assistant', content: systemPrompt },
|
||
{ role: 'user', content: userPrompt },
|
||
],
|
||
thinking: { type: 'disabled' },
|
||
});
|
||
|
||
const content = completion.choices[0]?.message?.content || '';
|
||
return content;
|
||
}
|
||
|
||
function scriptToSegments(script: string, hostName: string, guestName: string): Segment[] {
|
||
const segments: Segment[] = [];
|
||
const lines = script.split('\n');
|
||
|
||
let current: Segment | null = null;
|
||
let idx = 0;
|
||
|
||
const hostPrefix = `**${hostName}**：`;
|
||
const guestPrefix = `**${guestName}**：`;
|
||
|
||
for (const rawLine of lines) {
|
||
const line = rawLine.trim();
|
||
if (!line) continue;
|
||
|
||
if (line.startsWith(hostPrefix)) {
|
||
idx++;
|
||
current = {
|
||
idx,
|
||
speaker: 'host',
|
||
name: hostName,
|
||
text: line.slice(hostPrefix.length).trim(),
|
||
};
|
||
segments.push(current);
|
||
} else if (line.startsWith(guestPrefix)) {
|
||
idx++;
|
||
current = {
|
||
idx,
|
||
speaker: 'guest',
|
||
name: guestName,
|
||
text: line.slice(guestPrefix.length).trim(),
|
||
};
|
||
segments.push(current);
|
||
} else {
|
||
if (current) {
|
||
current.text = (current.text + ' ' + line).trim();
|
||
}
|
||
}
|
||
}
|
||
|
||
return segments;
|
||
}
|
||
|
||
function validateScript(
|
||
script: string,
|
||
cfg: GenConfig,
|
||
budgetLow: number,
|
||
budgetHigh: number
|
||
): [boolean, string[]] {
|
||
const reasons: string[] = [];
|
||
|
||
if (cfg.mode === 'dual') {
|
||
const hostTag = `**${cfg.hostName}**：`;
|
||
const guestTag = `**${cfg.guestName}**：`;
|
||
|
||
if (!script.includes(hostTag)) reasons.push(`缺少主持人标识：${hostTag}`);
|
||
if (!script.includes(guestTag)) reasons.push(`缺少嘉宾标识：${guestTag}`);
|
||
|
||
const turns = script.split('\n').filter(line =>
|
||
line.startsWith(hostTag) || line.startsWith(guestTag)
|
||
);
|
||
if (turns.length < 8) reasons.push('对谈轮次过少：建议至少 8 轮');
|
||
} else {
|
||
const speakerName = cfg.mode === 'single-male' ? cfg.hostName : cfg.guestName;
|
||
const speakerTag = `**${speakerName}**：`;
|
||
|
||
if (!script.includes(speakerTag)) reasons.push(`缺少主播标识：${speakerTag}`);
|
||
|
||
const turns = script.split('\n').filter(line => line.startsWith(speakerTag));
|
||
if (turns.length < 5) reasons.push('播客段数过少：建议至少 5 段');
|
||
}
|
||
|
||
const n = countNonWsChars(script);
|
||
if (n < budgetLow || n > budgetHigh) {
|
||
reasons.push(`字数不在预算：当前约 ${n} 字，预算 ${budgetLow}-${budgetHigh}`);
|
||
}
|
||
|
||
// 只检查开场和总结，不检查"核心点1/2/3"标签（因为不应该出现在对话中）
|
||
const mustHave = ['开场', '总结'];
|
||
for (const kw of mustHave) {
|
||
if (!script.includes(kw)) {
|
||
reasons.push(`缺少结构要素：${kw}（请在对话中自然引入）`);
|
||
}
|
||
}
|
||
|
||
// 检查是否有足够的对话轮次（确保内容覆盖了多个主题）
|
||
const lineCount = script.split('\n').filter(l => l.trim()).length;
|
||
if (lineCount < 10) {
|
||
reasons.push('对话轮次过少，建议至少10段对话');
|
||
}
|
||
|
||
return [reasons.length === 0, reasons];
|
||
}
|
||
|
||
function makeRetryHint(reasons: string[], cfg: GenConfig, budgetLow: number, budgetHigh: number): string {
|
||
const lines = ['请严格修复以下问题后重新生成：'];
|
||
for (const r of reasons) lines.push(`- ${r}`);
|
||
lines.push(`- 总字数必须在 ${budgetLow}-${budgetHigh} 之间。`);
|
||
|
||
if (cfg.mode === 'dual') {
|
||
lines.push(`- 每段必须以"**${cfg.hostName}**："或"**${cfg.guestName}**："开头。`);
|
||
} else {
|
||
const speakerName = cfg.mode === 'single-male' ? cfg.hostName : cfg.guestName;
|
||
lines.push(`- 所有内容都由一人讲述，每段必须以"**${speakerName}**："开头。`);
|
||
}
|
||
|
||
lines.push('- 必须包含开场和总结，中间用自然过渡语连接不同主题，不要出现"核心点1/2/3"等标签。');
|
||
return lines.join('\n');
|
||
}
|
||
|
||
async function ttsRequest(
|
||
zai: any,
|
||
text: string,
|
||
voice: string,
|
||
speed: number
|
||
): Promise<Buffer> {
|
||
const response = await zai.audio.tts.create({
|
||
input: text,
|
||
voice: voice,
|
||
speed: speed,
|
||
response_format: 'wav',
|
||
stream: false,
|
||
});
|
||
|
||
const arrayBuffer = await response.arrayBuffer();
|
||
const buffer = Buffer.from(new Uint8Array(arrayBuffer));
|
||
return buffer;
|
||
}
|
||
|
||
function ensureSilenceWav(filePath: string, params: { nchannels: number; sampwidth: number; framerate: number }, ms: number): void {
|
||
const { nchannels, sampwidth, framerate } = params;
|
||
const nframes = Math.floor((framerate * ms) / 1000);
|
||
const silenceFrame = Buffer.alloc(sampwidth * nchannels, 0);
|
||
const frames = Buffer.alloc(silenceFrame.length * nframes, 0);
|
||
|
||
const header = Buffer.alloc(44);
|
||
header.write('RIFF', 0);
|
||
header.writeUInt32LE(36 + frames.length, 4);
|
||
header.write('WAVE', 8);
|
||
header.write('fmt ', 12);
|
||
header.writeUInt32LE(16, 16);
|
||
header.writeUInt16LE(1, 20);
|
||
header.writeUInt16LE(nchannels, 22);
|
||
header.writeUInt32LE(framerate, 24);
|
||
header.writeUInt32LE(framerate * nchannels * sampwidth, 28);
|
||
header.writeUInt16LE(nchannels * sampwidth, 32);
|
||
header.writeUInt16LE(sampwidth * 8, 34);
|
||
header.write('data', 36);
|
||
header.writeUInt32LE(frames.length, 40);
|
||
|
||
fs.writeFileSync(filePath, Buffer.concat([header, frames]));
|
||
}
|
||
|
||
function wavParams(filePath: string): { nchannels: number; sampwidth: number; framerate: number } {
|
||
const buffer = fs.readFileSync(filePath);
|
||
const nchannels = buffer.readUInt16LE(22);
|
||
const sampwidth = buffer.readUInt16LE(34) / 8;
|
||
const framerate = buffer.readUInt32LE(24);
|
||
return { nchannels, sampwidth, framerate };
|
||
}
|
||
|
||
function joinWavsWave(outPath: string, wavPaths: string[], pauseMs: number): void {
|
||
if (wavPaths.length === 0) throw new Error('No wav files to join.');
|
||
|
||
const ref = wavPaths[0];
|
||
const refParams = wavParams(ref);
|
||
const silencePath = path.join(os.tmpdir(), `_silence_${Date.now()}.wav`);
|
||
if (pauseMs > 0) ensureSilenceWav(silencePath, refParams, pauseMs);
|
||
|
||
const chunks: Buffer[] = [];
|
||
|
||
for (let i = 0; i < wavPaths.length; i++) {
|
||
const wavPath = wavPaths[i];
|
||
const buffer = fs.readFileSync(wavPath);
|
||
const dataStart = buffer.indexOf('data') + 8;
|
||
const data = buffer.subarray(dataStart);
|
||
|
||
const params = wavParams(wavPath);
|
||
if (params.nchannels !== refParams.nchannels ||
|
||
params.sampwidth !== refParams.sampwidth ||
|
||
params.framerate !== refParams.framerate) {
|
||
throw new Error(`WAV params mismatch: ${wavPath}`);
|
||
}
|
||
|
||
chunks.push(data);
|
||
|
||
if (pauseMs > 0 && i < wavPaths.length - 1) {
|
||
const silenceBuffer = fs.readFileSync(silencePath);
|
||
const silenceData = silenceBuffer.subarray(silenceBuffer.indexOf('data') + 8);
|
||
chunks.push(silenceData);
|
||
}
|
||
}
|
||
|
||
const totalDataSize = chunks.reduce((sum, buf) => sum + buf.length, 0);
|
||
const header = Buffer.alloc(44);
|
||
header.write('RIFF', 0);
|
||
header.writeUInt32LE(36 + totalDataSize, 4);
|
||
header.write('WAVE', 8);
|
||
header.write('fmt ', 12);
|
||
header.writeUInt32LE(16, 16);
|
||
header.writeUInt16LE(1, 20);
|
||
header.writeUInt16LE(refParams.nchannels, 22);
|
||
header.writeUInt32LE(refParams.framerate, 24);
|
||
header.writeUInt32LE(refParams.framerate * refParams.nchannels * refParams.sampwidth, 28);
|
||
header.writeUInt16LE(refParams.nchannels * refParams.sampwidth, 32);
|
||
header.writeUInt16LE(refParams.sampwidth * 8, 34);
|
||
header.write('data', 36);
|
||
header.writeUInt32LE(totalDataSize, 40);
|
||
|
||
const output = Buffer.concat([header, ...chunks]);
|
||
fs.writeFileSync(outPath, output);
|
||
|
||
if (fs.existsSync(silencePath)) fs.unlinkSync(silencePath);
|
||
}
|
||
|
||
// -----------------------------
|
||
// Main
|
||
// -----------------------------
|
||
async function main() {
|
||
const args = parseArgs();
|
||
|
||
const inputPath = args.input;
|
||
const outDir = args.out_dir;
|
||
const topic = args.topic;
|
||
|
||
// 检查参数：必须提供 input 或 topic 之一
|
||
if ((!inputPath && !topic) || !outDir) {
|
||
console.error('Usage: tsx generate.ts --input=<file> --out_dir=<dir>');
|
||
console.error(' OR: tsx generate.ts --topic=<search-term> --out_dir=<dir>');
|
||
console.error('');
|
||
console.error('Examples:');
|
||
console.error(' # From file');
|
||
console.error(' npm run generate -- --input=article.txt --out_dir=out');
|
||
console.error(' # From web search');
|
||
console.error(' npm run generate -- --topic="最新AI新闻" --out_dir=out');
|
||
process.exit(1);
|
||
}
|
||
|
||
// Merge config
|
||
const cfg: GenConfig = {
|
||
...DEFAULT_CONFIG,
|
||
mode: (args.mode || 'dual') as GenConfig['mode'],
|
||
durationManual: parseInt(args.duration || '0'),
|
||
hostName: args.host_name || DEFAULT_CONFIG.hostName,
|
||
guestName: args.guest_name || DEFAULT_CONFIG.guestName,
|
||
voiceHost: args.voice_host || DEFAULT_CONFIG.voiceHost,
|
||
voiceGuest: args.voice_guest || DEFAULT_CONFIG.voiceGuest,
|
||
speed: parseFloat(args.speed || String(DEFAULT_CONFIG.speed)),
|
||
pauseMs: parseInt(args.pause_ms || String(DEFAULT_CONFIG.pauseMs)),
|
||
};
|
||
|
||
// Create output directory
|
||
if (!fs.existsSync(outDir)) {
|
||
fs.mkdirSync(outDir, { recursive: true });
|
||
}
|
||
|
||
// 根据模式获取资料
|
||
let material: string;
|
||
let inputSource: string;
|
||
|
||
if (inputPath) {
|
||
// 模式1：从文件读取
|
||
console.log(`[MODE] Reading from file: ${inputPath}`);
|
||
material = readText(inputPath);
|
||
inputSource = `file:${inputPath}`;
|
||
} else if (topic) {
|
||
// 模式2：联网搜索
|
||
console.log(`[MODE] Searching web for topic: ${topic}`);
|
||
const zai = await ZAI.create();
|
||
|
||
const searchResults = await zai.functions.invoke('web_search', {
|
||
query: topic,
|
||
num: 10
|
||
});
|
||
|
||
if (!Array.isArray(searchResults) || searchResults.length === 0) {
|
||
console.error(`未找到关于"${topic}"的搜索结果`);
|
||
process.exit(2);
|
||
}
|
||
|
||
console.log(`[SEARCH] Found ${searchResults.length} results`);
|
||
|
||
// 将搜索结果转换为文本资料
|
||
material = searchResults
|
||
.map((r: any, i: number) => `【来源 ${i + 1}】${r.name}\n${r.snippet}\n链接：${r.url}`)
|
||
.join('\n\n');
|
||
|
||
inputSource = `web_search:${topic}`;
|
||
console.log(`[SEARCH] Compiled material (${material.length} chars)`);
|
||
} else {
|
||
console.error('[ERROR] Neither --input nor --topic provided');
|
||
process.exit(1);
|
||
}
|
||
|
||
const inputChars = material.length;
|
||
|
||
// Calculate duration
|
||
let durationMin: number;
|
||
if (cfg.durationManual >= 3 && cfg.durationManual <= 20) {
|
||
durationMin = cfg.durationManual;
|
||
} else {
|
||
durationMin = chooseDurationMinutes(inputChars, DURATION_RANGE_LOW, DURATION_RANGE_HIGH);
|
||
}
|
||
|
||
const [target, low, high] = charBudget(durationMin, cfg.charsPerMin, BUDGET_TOLERANCE);
|
||
|
||
console.log(`[INFO] input_chars=${inputChars} duration=${durationMin}min budget=${low}-${high}`);
|
||
|
||
let attemptHint = '';
|
||
let lastScript: string | null = null;
|
||
|
||
// Initialize ZAI SDK (reuse for TTS)
|
||
const zai = await ZAI.create();
|
||
|
||
// Generate script
|
||
for (let attempt = 1; attempt <= cfg.maxAttempts; attempt++) {
|
||
const [systemPrompt, userPrompt] = buildPrompts(
|
||
material,
|
||
cfg,
|
||
durationMin,
|
||
target,
|
||
low,
|
||
high,
|
||
attemptHint
|
||
);
|
||
|
||
try {
|
||
console.log(`[LLM] Attempt ${attempt}/${cfg.maxAttempts}...`);
|
||
const content = await callZAI(systemPrompt, userPrompt, cfg.temperature);
|
||
lastScript = content;
|
||
|
||
const [ok, reasons] = validateScript(content, cfg, low, high);
|
||
|
||
if (ok) {
|
||
break;
|
||
}
|
||
|
||
attemptHint = makeRetryHint(reasons, cfg, low, high);
|
||
console.error(`[WARN] Validation failed:`, reasons.join(', '));
|
||
} catch (error: any) {
|
||
console.error(`[ERROR] LLM call failed: ${error.message}`);
|
||
throw error;
|
||
}
|
||
}
|
||
|
||
if (!lastScript) {
|
||
console.error('[ERROR] 未生成任何脚本输出。');
|
||
process.exit(1);
|
||
}
|
||
|
||
// Write script
|
||
const scriptPath = path.join(outDir, 'podcast_script.md');
|
||
fs.writeFileSync(scriptPath, lastScript, 'utf-8');
|
||
console.log(`[DONE] podcast_script.md -> ${scriptPath}`);
|
||
|
||
// Parse segments
|
||
const segments = scriptToSegments(lastScript, cfg.hostName, cfg.guestName);
|
||
console.log(`[INFO] Parsed ${segments.length} segments`);
|
||
|
||
// Generate TTS using SDK
|
||
const tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'podcast_segments_'));
|
||
const produced: string[] = [];
|
||
|
||
try {
|
||
for (let i = 0; i < segments.length; i++) {
|
||
const seg = segments[i];
|
||
const text = seg.text.trim();
|
||
if (!text) continue;
|
||
|
||
let voice: string;
|
||
if (cfg.mode === 'dual') {
|
||
voice = seg.speaker === 'host' ? cfg.voiceHost : cfg.voiceGuest;
|
||
} else if (cfg.mode === 'single-male') {
|
||
voice = cfg.voiceHost;
|
||
} else {
|
||
voice = cfg.voiceGuest;
|
||
}
|
||
|
||
const wavPath = path.join(tmpDir, `seg_${seg.idx.toString().padStart(4, '0')}.wav`);
|
||
|
||
console.log(`[TTS] [${i + 1}/${segments.length}] idx=${seg.idx} speaker=${seg.speaker} voice=${voice}`);
|
||
|
||
const buffer = await ttsRequest(zai, text, voice, cfg.speed);
|
||
fs.writeFileSync(wavPath, buffer);
|
||
produced.push(wavPath);
|
||
}
|
||
|
||
// Join segments
|
||
const podcastPath = path.join(outDir, 'podcast.wav');
|
||
console.log(`[JOIN] Joining ${produced.length} wav files -> ${podcastPath}`);
|
||
|
||
joinWavsWave(podcastPath, produced, cfg.pauseMs);
|
||
console.log(`[DONE] podcast.wav -> ${podcastPath}`);
|
||
|
||
} finally {
|
||
// Cleanup temp directory
|
||
try {
|
||
fs.rmSync(tmpDir, { recursive: true, force: true });
|
||
} catch (error: any) {
|
||
console.error(`[WARN] Failed to cleanup temp dir: ${error.message}`);
|
||
}
|
||
}
|
||
|
||
console.log('\n[FINAL OUTPUT]');
|
||
console.log(` 📄 podcast_script.md -> ${scriptPath}`);
|
||
console.log(` 🎙️ podcast.wav -> ${path.join(outDir, 'podcast.wav')}`);
|
||
}
|
||
|
||
main().catch(error => {
|
||
console.error('[FATAL ERROR]', error);
|
||
process.exit(1);
|
||
});
|