Aliyun Cosyvoice Voice Clone

interface:
  display_name: "Alibaba Cloud AI Audio CosyVoice Voice Clone"
  short_description: "CosyVoice custom voice enrollment from reference audio"
  default_prompt: "Use $aliyun-cosyvoice-voice-clone to complete this ai/audio CosyVoice voice clone task on Alibaba Cloud."

{
  "skill_name": "alicloud-ai-audio-cosyvoice-voice-clone",
  "evals": [
    {
      "id": 1,
      "prompt": "用阿里云百炼的 CosyVoice 给我做一个声音复刻。参考音频是 https://example.com/female-host.wav，目标模型用 cosyvoice-v3.5-plus，前缀就叫 host01。请给我可复用的请求 JSON，并说明后续合成时 voice_id 要怎么用。",
      "expected_output": "Produces a clone enrollment request for cosyvoice-v3.5-plus and explains that the returned voice_id must be reused in later TTS calls with the same target_model.",
      "files": [],
      "expectations": [
        "The output uses model \"voice-enrollment\".",
        "The output sets target_model to cosyvoice-v3.5-plus.",
        "The output includes prefix and reference audio URL.",
        "The output explains that voice_id is reused in later synthesis requests."
      ]
    },
    {
      "id": 2,
      "prompt": "我要给客服机器人做英文音色克隆，音频地址是 https://example.com/agent-en.wav，语言提示 en，前缀 agent7。帮我准备最小可用的 CosyVoice clone 请求，并提醒我有哪些区域限制。",
      "expected_output": "Produces a minimal English clone request and notes that cosyvoice-v3.5-plus/flash clone capabilities are limited to China mainland deployment mode.",
      "files": [],
      "expectations": [
        "The request contains language_hints with en.",
        "The request includes a public URL field for the enrollment sample.",
        "The answer mentions China mainland deployment restrictions for v3.5 clone/design."
      ]
    },
    {
      "id": 3,
      "prompt": "帮我生成一个 CosyVoice 声音复刻模板，我要控制 max_prompt_audio_length=12 和 enable_preprocess=true，模型还是 cosyvoice-v3.5-plus，样本地址 https://example.com/news.wav，前缀 news9。",
      "expected_output": "Produces a clone request including the optional preprocessing fields supported by cosyvoice-v3.5-plus.",
      "files": [],
      "expectations": [
        "The request includes max_prompt_audio_length set to 12.",
        "The request includes enable_preprocess set to true.",
        "The request keeps target_model as cosyvoice-v3.5-plus."
      ]
    }
  ]
}

#!/usr/bin/env python3
"""Prepare and optionally validate a CosyVoice clone enrollment request."""

from __future__ import annotations

import argparse
import json
import sys
from pathlib import Path


def _load_json(path: str) -> dict:
    return json.loads(Path(path).read_text(encoding="utf-8"))


def main() -> None:
    parser = argparse.ArgumentParser(description="Prepare CosyVoice voice clone request")
    parser.add_argument("--target-model", default="cosyvoice-v3.5-plus")
    parser.add_argument("--prefix", required=True)
    parser.add_argument("--voice-sample-url", required=True)
    parser.add_argument("--language-hint", default="zh")
    parser.add_argument("--max-prompt-audio-length", type=float)
    parser.add_argument("--enable-preprocess", action="store_true")
    parser.add_argument(
        "--output",
        default="output/aliyun-cosyvoice-voice-clone/request.json",
    )
    parser.add_argument("--validate-response", help="Path to a JSON response to validate")
    args = parser.parse_args()

    request = {
        "model": "voice-enrollment",
        "input": {
            "action": "create_voice",
            "target_model": args.target_model,
            "prefix": args.prefix,
            "url": args.voice_sample_url,
            "language_hints": [args.language_hint],
        },
    }

    if args.max_prompt_audio_length is not None:
        request["input"]["max_prompt_audio_length"] = args.max_prompt_audio_length
    if args.enable_preprocess:
        request["input"]["enable_preprocess"] = True

    out = Path(args.output)
    out.parent.mkdir(parents=True, exist_ok=True)
    out.write_text(json.dumps(request, ensure_ascii=False, indent=2), encoding="utf-8")

    result = {"ok": True, "request_path": str(out)}
    if args.validate_response:
        response = _load_json(args.validate_response)
        voice_id = ((response.get("output") or {}).get("voice_id"))
        if not voice_id:
            print(json.dumps({"ok": False, "error": "missing output.voice_id"}, ensure_ascii=False))
            sys.exit(1)
        result["response_valid"] = True

    print(json.dumps(result, ensure_ascii=False))


if __name__ == "__main__":
    main()