大模型服务平台百炼：声音复刻

Qwen-Audio-TTS / CosyVoice

Qwen-Audio-TTS 和 CosyVoice 使用相同的音色创建和基础合成流程。以下以 qwen-audio-3.1-tts-flash 为例。

1. 创建音色：通过 HTTP API 的 audio 字段传入参考音频 URL 或 Base64 Data URI，target_model 指定目标模型，prefix 为音色名称前缀。
2. 合成语音：将返回的音色 ID 传入 voice，使用同一个目标模型合成语音并保存音频。

import os
import time
import requests

workspace_id = "YOUR_WORKSPACE_ID"
target_model = "qwen-audio-3.1-tts-flash"
base_url = f"https://{workspace_id}.cn-beijing.maas.aliyuncs.com/api/v1"
headers = {"Authorization": f"Bearer {os.environ['DASHSCOPE_API_KEY']}"}

response = requests.post(
    f"{base_url}/services/audio/tts/customization",
    headers=headers,
    json={
        "model": "voice-enrollment",
        "input": {
            "action": "create_voice",
            "target_model": target_model,
            "prefix": "myvoice",
            "audio": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/cosyvoice/cosyvoice-zeroshot-sample.wav",
        },
    },
    timeout=60,
)
response.raise_for_status()
voice_id = response.json()["output"]["voice_id"]
print(f"voice_id: {voice_id}")

for _ in range(30):
    response = requests.post(
        f"{base_url}/services/audio/tts/customization",
        headers=headers,
        json={
            "model": "voice-enrollment",
            "input": {"action": "query_voice", "voice_id": voice_id},
        },
        timeout=60,
    )
    response.raise_for_status()
    status = response.json()["output"]["status"]
    if status == "OK":
        break
    if status != "DEPLOYING":
        raise RuntimeError(f"Voice status: {status}")
    time.sleep(1)
else:
    raise TimeoutError("Voice is not ready")

response = requests.post(
    f"{base_url}/services/audio/tts/SpeechSynthesizer",
    headers=headers,
    json={
        "model": target_model,
        "input": {
            "text": "今天天气晴朗，适合外出散步。祝你度过愉快的一天。",
            "voice": voice_id,
            "format": "wav",
            "sample_rate": 24000,
        },
    },
    timeout=60,
)
response.raise_for_status()
audio_url = response.json()["output"]["audio"]["url"]
audio_response = requests.get(audio_url, timeout=60)
audio_response.raise_for_status()
with open("output.wav", "wb") as f:
    f.write(audio_response.content)

运行后，合成音频保存为 output.wav。保存打印的音色 ID，后续可继续使用。

Qwen-Audio-Realtime

Qwen-Audio-Realtime 的复刻音色用于自定义实时语音对话的回复声音。以下以 qwen-audio-3.1-realtime-plus 为例创建音色。

import os
import time
import requests

workspace_id = "YOUR_WORKSPACE_ID"
target_model = "qwen-audio-3.1-realtime-plus"
base_url = f"https://{workspace_id}.cn-beijing.maas.aliyuncs.com/api/v1"
headers = {"Authorization": f"Bearer {os.environ['DASHSCOPE_API_KEY']}"}

response = requests.post(
    f"{base_url}/services/audio/tts/customization",
    headers=headers,
    json={
        "model": "voice-enrollment",
        "input": {
            "action": "create_voice",
            "target_model": target_model,
            "prefix": "myvoice",
            "audio": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/cosyvoice/cosyvoice-zeroshot-sample.wav",
        },
    },
    timeout=60,
)
response.raise_for_status()
voice_id = response.json()["output"]["voice_id"]
print(f"voice_id: {voice_id}")

for _ in range(30):
    response = requests.post(
        f"{base_url}/services/audio/tts/customization",
        headers=headers,
        json={
            "model": "voice-enrollment",
            "input": {"action": "query_voice", "voice_id": voice_id},
        },
        timeout=60,
    )
    response.raise_for_status()
    status = response.json()["output"]["status"]
    if status == "OK":
        break
    if status != "DEPLOYING":
        raise RuntimeError(f"Voice status: {status}")
    time.sleep(1)
else:
    raise TimeoutError("Voice is not ready")

建立实时语音对话连接后，在首次 session.update 事件中，将 session.voice 设置为返回的音色 ID（替换下方的 VOICE_ID）。会话模型必须与创建音色时的 target_model 一致。

{
    "type": "session.update",
    "session": {
        "voice": "VOICE_ID"
    }
}

华北2（北京）

使用该地域的 API Key 调用以下模型：

• Qwen-Audio-TTS：qwen-audio-3.1-tts-plus、qwen-audio-3.1-tts-flash、qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash

• Qwen-Audio-Realtime：qwen-audio-3.1-realtime-plus、qwen-audio-3.0-realtime-plus、qwen-audio-3.0-realtime-flash

• CosyVoice：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-plus、cosyvoice-v3-flash、cosyvoice-v2

新加坡

使用该地域的 API Key 调用以下模型：

• Qwen-Audio-TTS：qwen-audio-3.1-tts-plus、qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash

• CosyVoice：cosyvoice-v3-plus、cosyvoice-v3-flash