wx_pbzc/cloudfunctions/tts/index.js

// Cloud function: synthesize the 4 fixed training-voice prompts.
//
// We do NOT cache on the server. Two reasons:
//   1. Caching via cloud storage requires self-constructing a fileID of
//      the form `cloud://<full-env-id>/...`, but `getWXContext().ENV`
//      only returns the simple env id. Mismatched → unopenable fileID.
//   2. Caching via the cloud database requires `cloud.database()` at
//      top level, which on this WeChat cloud runtime crashes the SCF
//      framework's `writeRuntimeFile` step (it can't toString() the
//      error object the SDK throws).
//
// The client (`utils/voice.js`) caches audioUrls in memory per session,
// so each of the 4 prompts is synthesized at most once per app launch.
// Re-synthesizing on app restart is fine — 800万 chars/month free tier
// covers 4 × ~30 chars × thousands of restarts per month.
//
// Configuration (云开发控制台 → 云函数 → tts → 函数配置 → 环境变量):
//   TTS_SECRET_ID   — Tencent Cloud API key id
//   TTS_SECRET_KEY  — Tencent Cloud API key
//   TTS_REGION      — (optional) defaults to ap-guangzhou
//
// NOTE: env var names cannot start with SCF_ / QCLOUD_ / TENCENTCLOUD_ —
// those prefixes are reserved by Tencent Cloud SCF. Use the TTS_ prefix.

const cloud = require('wx-server-sdk')
cloud.init()

// Polished, fixed prompts.
const PROMPTS = {
  halfway: '已完成一半啦,坚持就是胜利!',
  last30: '最后30秒,保持呼吸,稳住姿势!',
  last10: '最后10秒,再加把劲!',
  complete: '太棒了!今天的目标已完成,继续加油!'
}

const CACHE_DIR = 'tts-cache'

const _synthesize = async (text, promptKey) => {
  let TtsClient
  try {
    TtsClient = require('tencentcloud-sdk-nodejs').tts.v20190823.Client
  } catch (e) {
    throw new Error('tencentcloud-sdk-nodejs not installed. Run `npm i tencentcloud-sdk-nodejs` in this cloud function directory.')
  }

  const secretId = process.env.TTS_SECRET_ID
  const secretKey = process.env.TTS_SECRET_KEY
  if (!secretId || !secretKey) {
    throw new Error('TTS_SECRET_ID / TTS_SECRET_KEY not configured.')
  }

  const client = new TtsClient({
    credential: { secretId, secretKey },
    region: process.env.TTS_REGION || 'ap-guangzhou'
  })

  const res = await client.TextToVoice({
    Text: text,
    SessionId: `${promptKey}-${Date.now()}-${Math.random().toString(36).slice(2, 10)}`,
    VoiceType: 101001,    // 智瑜,温柔女声
    Codec: 'mp3',
    SampleRate: 16000,
    Speed: 0,
    Volume: 5
  })
  return Buffer.from(res.Audio, 'base64')
}

const _upload = async (key, buffer) => {
  const cloudPath = `${CACHE_DIR}/${key}.mp3`
  // uploadFile returns the canonical fileID (e.g. "cloud://env.xxx/...").
  // We must use THAT — not the relative cloudPath we passed in — when
  // calling getTempFileURL. A self-constructed fileID is invalid.
  const res = await cloud.uploadFile({ cloudPath, fileContent: buffer })
  return res.fileID
}

exports.main = async (event) => {
  const { promptKey } = event || {}
  const text = PROMPTS[promptKey]
  if (!text) {
    return { success: false, error: `Unknown prompt key: ${promptKey}` }
  }

  try {
    const buffer = await _synthesize(text, promptKey)
    const newFileID = await _upload(promptKey, buffer)
    const urlRes = await cloud.getTempFileURL({ fileList: [newFileID] })
    const url = urlRes.fileList[0].tempFileURL
    if (!url) {
      return { success: false, error: 'getTempFileURL returned empty URL', text }
    }
    return { success: true, audioUrl: url, text }
  } catch (e) {
    return { success: false, error: e.message, text }
  }
}