ai-chatbot/lib/video-canvas/audio-contract.ts

import { z } from "zod";
const qwenVoices = [
  { id: "Cherry", name: "芊悦" },
  { id: "Serena", name: "苏瑶" },
];
export const audioModelContracts = {
  "qwen3-tts-instruct-flash": {
    kind: "tts",
    label: "千问3 TTS 指令版",
    voices: qwenVoices,
    instructions: true,
    controls: false,
    formats: ["wav"],
  },
  "qwen3-tts-flash": {
    kind: "tts",
    label: "千问3 TTS",
    voices: [...qwenVoices, { id: "Ryan", name: "甜茶" }, { id: "Jennifer", name: "詹妮弗" }],
    instructions: false,
    controls: false,
    formats: ["wav"],
  },
  "qwen-audio-3.0-tts-flash": {
    kind: "tts",
    label: "千问 Audio 3.0 Flash",
    voices: [
      { id: "longanhuan_v3.6", name: "龙安欢" },
      { id: "longanfengyue", name: "龙安风悦" },
      { id: "longanyuanfei", name: "龙安元妃" },
      { id: "longanlingxi", name: "龙安灵希" },
    ],
    instructions: true,
    controls: true,
    formats: ["mp3", "wav"],
  },
  "qwen-audio-3.0-tts-plus": {
    kind: "tts",
    label: "千问 Audio 3.0 Plus",
    voices: [
      { id: "longanlingxin", name: "龙安灵心" },
      { id: "longanlufeng", name: "龙安鲁风" },
    ],
    instructions: true,
    controls: true,
    formats: ["mp3", "wav"],
  },
  "fun-music-v1": {
    kind: "music",
    label: "Fun-Music 音乐",
    voices: [],
    instructions: false,
    controls: false,
    formats: ["mp3", "wav"],
  },
} satisfies Record<
  string,
  {
    kind: string;
    label: string;
    voices: { id: string; name: string }[];
    instructions: boolean;
    controls: boolean;
    formats: string[];
  }
>;
export type AudioModelId = keyof typeof audioModelContracts;
export const audioRequestSchema = z.object({
  mode: z.enum(["tts", "music"]),
  modelRef: z.string().min(1),
  text: z.string().trim().max(2000).default(""),
  lyrics: z.string().trim().max(2000).default(""),
  voice: z.string().max(100).optional(),
  language: z
    .enum([
      "Auto",
      "Chinese",
      "English",
      "German",
      "Italian",
      "Portuguese",
      "Spanish",
      "Japanese",
      "Korean",
      "French",
      "Russian",
    ])
    .default("Chinese"),
  instructions: z.string().trim().max(1200).default(""),
  format: z.enum(["mp3", "wav"]).default("wav"),
  instrumental: z.boolean().default(true),
  gender: z.enum(["male", "female"]).default("female"),
  rate: z.number().min(0.5).max(2).default(1),
  pitch: z.number().min(0.5).max(2).default(1),
  volume: z.number().int().min(0).max(100).default(50),
});
export type AudioRequest = z.infer<typeof audioRequestSchema>;
export function audioContract(id: string) {
  return Object.hasOwn(audioModelContracts, id)
    ? audioModelContracts[id as AudioModelId]
    : undefined;
}
export function officialAudioBase(base: string, modelId?: string) {
  const url = new URL(base);
  if (
    url.protocol !== "https:" ||
    url.username ||
    url.password ||
    !/^(?:dashscope(?:-intl|-us)?\.aliyuncs\.com|[a-zA-Z0-9-]+\.(?:cn-beijing|ap-southeast-1)\.maas\.aliyuncs\.com)$/.test(
      url.hostname,
    )
  )
    throw Error("请选择阿里云官方音频服务线路");
  if (
    (modelId === "fun-music-v1" || modelId?.startsWith("qwen-audio-")) &&
    url.hostname !== "dashscope.aliyuncs.com" &&
    !url.hostname.endsWith(".cn-beijing.maas.aliyuncs.com")
  )
    throw Error("所选模型仅支持阿里云北京地域");
  return url.origin;
}
export function buildAudioRequest(value: unknown, modelId: string) {
  const p = audioRequestSchema.parse(value),
    contract = audioContract(modelId);
  if (!contract || contract.kind !== p.mode) throw Error("音频模式与模型不匹配");
  if (!contract.formats.includes(p.format)) throw Error("所选模型不支持这个音频格式");
  if (p.mode === "music") {
    if ((p.instrumental && !p.text) || (!p.text && !p.lyrics)) throw Error("请输入音乐描述或歌词");
    if (
      !p.instrumental &&
      p.lyrics &&
      (p.lyrics.length < 5 || (/[\u3400-\u9fff]/.test(p.lyrics) && p.lyrics.length > 350))
    )
      throw Error("中文歌词需 5–350 字，英文歌词需 5–2000 字");
    return {
      path: "/api/v1/services/audio/music/generation",
      body: {
        model: modelId,
        input: {
          ...(p.text ? { prompt: p.text } : {}),
          ...(!p.instrumental && p.lyrics ? { lyrics: p.lyrics } : {}),
          is_instrumental: p.instrumental,
          ...(!p.instrumental ? { gender: p.gender } : {}),
          format: p.format,
        },
      },
    };
  }
  if (!p.text || p.text.length > 600) throw Error("配音文本需 1–600 字");
  if (!contract.voices.some((v) => v.id === p.voice)) throw Error("请选择当前模型支持的音色");
  if (!contract.instructions && p.instructions) throw Error("所选模型不支持风格指令");
  if (!contract.controls && (p.rate !== 1 || p.pitch !== 1 || p.volume !== 50))
    throw Error("所选模型不支持数值语速、音调和音量参数，可改用风格指令");
  const modern = modelId.startsWith("qwen-audio-");
  if (modern && !["Chinese", "English", "Auto"].includes(p.language))
    throw Error("当前系统音色支持中文和英文");
  return {
    path: modern
      ? "/api/v1/services/audio/tts/SpeechSynthesizer"
      : "/api/v1/services/aigc/multimodal-generation/generation",
    body: {
      model: modelId,
      input: {
        text: p.text,
        voice: p.voice,
        ...(modern
          ? {
              format: p.format,
              rate: p.rate,
              pitch: p.pitch,
              volume: p.volume,
              ...(p.instructions ? { instruction: p.instructions } : {}),
              ...(p.language !== "Auto"
                ? { language_hints: [p.language === "Chinese" ? "zh" : "en"] }
                : {}),
            }
          : {
              language_type: p.language,
              ...(p.instructions
                ? { instructions: p.instructions, optimize_instructions: true }
                : {}),
            }),
      },
    },
  };
}
export const audioCheckpointSchema = z.object({
  operation: z.literal("audio"),
  modelRef: z.string(),
  modelId: z.string(),
  url: z
    .string()
    .url()
    .refine((v) => /^https?:\/\//i.test(v)),
  format: z.enum(["mp3", "wav"]),
  requestId: z.string().optional(),
  expiresAt: z.number().optional(),
});