96 lines
4.9 KiB
TypeScript
96 lines
4.9 KiB
TypeScript
import fs from 'node:fs/promises'
|
||
import path from 'node:path'
|
||
import { createHash, randomUUID } from 'node:crypto'
|
||
import { STORAGE_ROOT } from '../utils/paths.js'
|
||
import { ffmpeg, checkFfmpegSuite } from '../utils/ffmpeg.js'
|
||
|
||
const MAX_BYTES = 15 * 1024 * 1024
|
||
const INLINE_PROVIDERS = new Set(['volcengine', 'minimax'])
|
||
|
||
function localAudioPath(url: string): string | null {
|
||
if (!/^\/?static\//.test(url)) return null
|
||
const file = path.resolve(STORAGE_ROOT, url.replace(/^\/?static\//, ''))
|
||
if (!file.startsWith(path.resolve(STORAGE_ROOT) + path.sep)) throw new Error('无效的本地音色路径')
|
||
return file
|
||
}
|
||
|
||
function publicAudioUrl(url: string): string {
|
||
const base = (process.env.PUBLIC_BASE_URL || '').trim().replace(/\/+$/, '')
|
||
if (!/^https?:\/\//.test(base)) {
|
||
throw new Error('当前视频模型需要公网音色地址。请配置后端 PUBLIC_BASE_URL,或切换至支持本地音色的 Seedance 2.0 / MiniMax H3。')
|
||
}
|
||
return `${base}/${url.replace(/^\//, '')}`
|
||
}
|
||
|
||
/** Validate before creating a paid task. M4A/AAC samples are converted without trimming the voice. */
|
||
export async function prepareReferenceAudio(refs: string[] | undefined, provider: string): Promise<string[]> {
|
||
const urls = [...new Set((refs || []).map(url => url.trim()).filter(Boolean))]
|
||
const minSeconds = provider === 'aliyun' ? 1 : 2
|
||
let durationSum = 0
|
||
const prepared: string[] = []
|
||
for (const url of urls) {
|
||
const file = localAudioPath(url)
|
||
if (!file) { prepared.push(url); continue }
|
||
if (!INLINE_PROVIDERS.has(provider)) publicAudioUrl(url)
|
||
let size: number
|
||
try { size = (await fs.stat(file)).size } catch { throw new Error('角色音色文件不存在,请重新上传音色') }
|
||
const ext = path.extname(file).toLowerCase()
|
||
if (!['.mp3', '.wav', '.m4a', '.aac'].includes(ext)) throw new Error('参考音色仅支持 MP3、WAV、M4A 或 AAC')
|
||
if (size > 20 * 1024 * 1024) throw new Error('音色文件过大,请上传不超过 20MB 的音频')
|
||
const suite = await checkFfmpegSuite()
|
||
if (!suite.ffprobe) throw new Error('音色校验需要 FFprobe,请修复后端 FFprobe 配置后重试')
|
||
const duration = await new Promise<number>((resolve, reject) => {
|
||
ffmpeg.ffprobe(file, (error, metadata) => {
|
||
if (error || !metadata.streams.some(stream => stream.codec_type === 'audio')) {
|
||
reject(new Error('无法读取音色文件,请上传有效的人声音频')); return
|
||
}
|
||
resolve(Number(metadata.format.duration))
|
||
})
|
||
})
|
||
if (!Number.isFinite(duration) || duration < minSeconds || duration > 15) {
|
||
throw new Error(`当前模型要求每段参考音色为 ${minSeconds}–15 秒,请裁剪或重新上传`)
|
||
}
|
||
durationSum += duration
|
||
if (durationSum > 15.05) throw new Error('本次引用的音色总时长不能超过 15 秒,请减少引用或缩短音色样本')
|
||
if (ext === '.wav' || ext === '.mp3') {
|
||
if (size > MAX_BYTES) throw new Error('当前模型要求每段参考音色不超过 15MB,请压缩或重新上传')
|
||
prepared.push(url.replace(/^\//, ''))
|
||
continue
|
||
}
|
||
if (!suite.ffmpeg) throw new Error('M4A / AAC 音色需要 FFmpeg 转换,请修复配置或上传 WAV / MP3')
|
||
const hash = createHash('sha256').update(await fs.readFile(file)).digest('hex')
|
||
const relative = `static/voice-references/${hash}.wav`
|
||
const target = localAudioPath(relative)!
|
||
await fs.mkdir(path.dirname(target), { recursive: true })
|
||
try { await fs.access(target) } catch {
|
||
const temporary = `${target}.${randomUUID()}.wav`
|
||
try {
|
||
await new Promise<void>((resolve, reject) => {
|
||
ffmpeg(file).noVideo().audioChannels(1).audioFrequency(48000).audioCodec('pcm_s16le')
|
||
.format('wav').on('end', () => resolve()).on('error', () => reject(new Error('音色转换失败,请改用 WAV / MP3 文件'))).save(temporary)
|
||
})
|
||
await fs.rename(temporary, target)
|
||
} finally { await fs.rm(temporary, { force: true }) }
|
||
}
|
||
prepared.push(relative)
|
||
}
|
||
return prepared
|
||
}
|
||
|
||
/** Seedance 2.0 / MiniMax H3 accept audio data URLs; Wan requires a public/OSS URL.
|
||
* https://docs.volcengine.com/docs/ark/create-video-generation-task-api
|
||
* https://platform.minimax.cn/docs/api-reference/video-generation-v2-create
|
||
* https://help.aliyun.com/en/model-studio/wan3-video-generation-api-reference
|
||
*/
|
||
export async function resolveReferenceAudioUrls(refs: string[] | undefined, provider: string): Promise<string[]> {
|
||
return Promise.all((refs || []).map(async url => {
|
||
const file = localAudioPath(url)
|
||
if (!file) return url
|
||
if (!INLINE_PROVIDERS.has(provider)) return publicAudioUrl(url)
|
||
const bytes = await fs.readFile(file)
|
||
if (bytes.length > MAX_BYTES) throw new Error('参考音色不能超过 15MB')
|
||
const format = path.extname(file).toLowerCase() === '.mp3' ? 'mp3' : 'wav'
|
||
return `data:audio/${format};base64,${bytes.toString('base64')}`
|
||
}))
|
||
}
|