video-create/.claude/skills/video-from-script/scripts/qwen-tts.js

#!/usr/bin/env node

/**
 * CosyVoice TTS 批量语音合成脚本
 * 通过 WebSocket 调用阿里云 DashScope CosyVoice API
 *
 * 输入 JSON 文件格式:
 * {
 *   "segments": [
 *     {"id": 1, "text": "第一段文案"},
 *     {"id": 2, "text": "第二段文案"}
 *   ],
 *   "voice": "longanyang",       // 可选，覆盖 config
 *   "output_dir": "./audio"      // 可选，默认 ./audio
 * }
 *
 * 输出 JSON (stdout):
 * {
 *   "segments": [
 *     {"id": 1, "text": "...", "audio": "./audio/seg_001.mp3", "duration": 3.456}
 *   ]
 * }
 *
 * 也可作为模块调用:
 *   const { synthesize } = require('./qwen-tts')
 *   const { filePath, duration } = await synthesize('你好世界', { voice: 'longanyang' })
 */

const fs = require('fs')
const path = require('path')
const { execFileSync } = require('child_process')

const CONFIG_PATH = path.join(__dirname, '..', '..', 'config.json')

function loadConfig() {
  if (!fs.existsSync(CONFIG_PATH)) throw new Error(`config.json 不存在: ${CONFIG_PATH}`)
  return JSON.parse(fs.readFileSync(CONFIG_PATH, 'utf-8'))
}

/**
 * 解析音色：名称 → ID。如果是名称则查 ttsVoices 映射表，否则原样返回。
 */
function resolveVoice(voice, config) {
  if (!voice) return voice
  const voices = config.ttsVoices || {}
  return voices[voice] || voice
}

function getAudioDuration(filePath) {
  try {
    const out = execFileSync('ffprobe', [
      '-v', 'quiet', '-show_entries', 'format=duration',
      '-of', 'default=noprint_wrappers=1:nokey=1', filePath,
    ], { encoding: 'utf-8', timeout: 10000 })
    return parseFloat(out.trim())
  } catch {
    const stat = fs.statSync(filePath)
    return stat.size * 8 / 32000
  }
}

/**
 * 单段语音合成（CosyVoice WebSocket）
 * @param {string} text
 * @param {object} options - { voice, model, outputDir, id, instruction }
 * @returns {Promise<{filePath: string, duration: number}>}
 */
function synthesize(text, options = {}) {
  return new Promise((resolve, reject) => {
    const config = loadConfig()

    const apiKey = options.apiKey || config.ttsApiKey
    if (!apiKey) { reject(new Error('ttsApiKey 未配置')); return }

    const model = options.model || config.ttsModel || 'cosyvoice-v3-flash'
    const voice = resolveVoice(options.voice || config.ttsVoice, config) || 'longanyang'
    const instruction = options.instruction || config.ttsInstruction || ''
    const outputDir = options.outputDir || './audio'

    fs.mkdirSync(outputDir, { recursive: true })

    text = text.trimEnd()
    if (!/[。！？；，.!?…]$/.test(text)) text += '。'

    const id = options.id || 1
    const fileName = `seg_${String(id).padStart(3, '0')}.mp3`
    const filePath = path.resolve(outputDir, fileName)

    const wsUrl = 'wss://dashscope.aliyuncs.com/api-ws/v1/inference'
    const ws = new WebSocket(wsUrl, {
      headers: { Authorization: `bearer ${apiKey}` },
    })

    const taskId = `tts_${Date.now()}_${id}`
    const chunks = []
    let settled = false

    const timer = setTimeout(() => {
      if (!settled) { settled = true; ws.close(); reject(new Error('TTS 超时 (60s)')) }
    }, 60000)

    ws.addEventListener('open', () => {
      // Step 1: run-task — empty input, no text
      ws.send(JSON.stringify({
        header: {
          task_id: taskId,
          action: 'run-task',
          streaming: 'duplex',
        },
        payload: {
          task_group: 'audio',
          task: 'tts',
          function: 'SpeechSynthesizer',
          model,
          parameters: {
            voice,
            format: 'mp3',
            sample_rate: 24000,
            volume: 50,
            rate: options.rate || 1.15,
            pitch_rate: 1.0,
            text_type: 'PlainText',
            ...(instruction ? { instruction } : {}),
          },
          input: {},
        },
      }))
    })

    ws.addEventListener('message', async (event) => {
      if (typeof event.data !== 'string') {
        const buf = event.data instanceof Blob
          ? Buffer.from(await event.data.arrayBuffer())
          : Buffer.from(event.data)
        chunks.push(buf)
        return
      }
      try {
        const msg = JSON.parse(event.data)
        const evt = msg.header?.event

        if (evt === 'task-started') {
          // Step 2: continue-task — send text
          ws.send(JSON.stringify({
            header: {
              task_id: taskId,
              action: 'continue-task',
              streaming: 'duplex',
            },
            payload: {
              task_group: 'audio',
              task: 'tts',
              function: 'SpeechSynthesizer',
              model,
              input: { text },
            },
          }))

          // Step 3: finish-task
          ws.send(JSON.stringify({
            header: {
              task_id: taskId,
              action: 'finish-task',
              streaming: 'duplex',
            },
            payload: {
              task_group: 'audio',
              task: 'tts',
              function: 'SpeechSynthesizer',
              input: {},
            },
          }))
        } else if (evt === 'task-finished') {
          clearTimeout(timer)
          ws.close()
          if (settled) return
          settled = true

          const audio = Buffer.concat(chunks)
          if (audio.length === 0) { reject(new Error('TTS 未返回音频')); return }

          fs.writeFileSync(filePath, audio)
          resolve({ filePath, duration: getAudioDuration(filePath) })
        } else if (evt === 'task-failed') {
          clearTimeout(timer)
          ws.close()
          if (settled) return
          settled = true
          reject(new Error(`TTS 失败: ${msg.header?.error_message || msg.header?.message || JSON.stringify(msg)}`))
        }
      } catch {}
    })

    ws.addEventListener('error', (e) => {
      clearTimeout(timer)
      if (!settled) { settled = true; reject(new Error(`WebSocket 错误: ${e.message || '连接失败'}`)) }
    })
  })
}

/**
 * 批量语音合成
 */
async function synthesizeBatch(segments, options = {}) {
  const results = []
  for (const seg of segments) {
    console.error(`  合成 #${seg.id}: ${seg.text.substring(0, 30)}...`)
    try {
      const { filePath, duration } = await synthesize(seg.text, {
        ...options,
        id: seg.id,
      })
      results.push({
        id: seg.id,
        text: seg.text,
        audio: filePath,
        duration: Math.round(duration * 1000) / 1000,
      })
    } catch (err) {
      results.push({
        id: seg.id,
        text: seg.text,
        audio: '',
        duration: 0,
        error: err.message,
      })
    }
    await new Promise(r => setTimeout(r, 500))
  }
  return results
}

// CLI 入口
async function main() {
  const inputJson = process.argv[2]
  if (!inputJson) {
    console.error('用法: node qwen-tts.js <input.json>')
    console.error('')
    console.error('input.json 格式:')
    console.error(JSON.stringify({
      segments: [{ id: 1, text: '文案' }],
      voice: 'longanyang',
      output_dir: './audio',
    }, null, 2))
    process.exit(1)
  }

  const config = JSON.parse(fs.readFileSync(inputJson, 'utf-8'))
  const results = await synthesizeBatch(config.segments, {
    voice: config.voice,
    outputDir: config.output_dir || './audio',
  })

  process.stdout.write(JSON.stringify({ segments: results }, null, 2) + '\n')
}

if (require.main === module) {
  main().catch(err => {
    console.error('TTS 合成失败:', err.message)
    process.exit(1)
  })
}

module.exports = { synthesize, synthesizeBatch }