免费咨询

STT模型 qwen3.8-omni-flash

STT模型 qwen3.8-omni-flash

qwen3.8-omni-flash 是通义千问全模态模型，具备强大的音频理解能力：通过文本提示词指定任务，可将语音精准转写为文字（STT），同时能描述环境声、音乐元素（如风格、乐器）、说话人情绪等音频内容，适用于语音转写、会议纪要、音频内容分析、安全审核、意图识别等多个领域。

📍 请求地址 

https://api.apexrouter.top/v1/chat/completions

🎯 模型名称

qwen3.8-omni-flash

💻 音频理解URL 调用示例 

# ============================================================================
#                    qwen3.8-omni-flash 音频理解示例 (URL方式)
# ============================================================================
# 功能说明：通过URL方式上传音频文件，使用qwen3.8-omni-flash模型进行音频内容理解
# 模型名称：qwen3.8-omni-flash
# ============================================================================

import requests

# ----------------------------------------------------------------------------
#                              API 配置信息
# ----------------------------------------------------------------------------
# 接口地址：知能API Chat Completions 端点
url = "https://api.apexrouter.top/v1/chat/completions"

# 请求头配置
headers = {
    "Authorization": "sk-*****************************************",  # API密钥
    "Content-Type": "application/json"                                              # 内容类型
}

# ----------------------------------------------------------------------------
#                              请求体构建
# ----------------------------------------------------------------------------
data = {
    # 指定使用的模型
    "model": "qwen3.8-omni-flash",

    # 消息列表
    "messages": [
        {
            "role": "user",
            "content": [
                {
                    # 文本提示词：指定音频处理任务（如转写为文字）
                    "type": "text",
                    "text": "请将这段音频转写为文字，并描述音频中的环境声音。"
                },
                {
                    # 音频输入类型
                    "type": "input_audio",
                    "input_audio": {
                        # 音频文件URL地址（支持公网可访问的音频链接）
                        "data": "https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20240916/xvappi/装修噪音.wav",
                        # 音频格式（支持 wav、mp3 等）
                        "format": "wav"
                    }
                }
            ]
        }
    ]
}

# ----------------------------------------------------------------------------
#                              发送请求并输出结果
# ----------------------------------------------------------------------------
response = requests.post(url, headers=headers, json=data)
print(response.json())

📥 返回示例 

{
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "message": {
        "content": "**音频转写：**\n哎呀，这样我还怎么安静工作啊\n\n**环境声音描述：**\n背景中有非常清晰且规律的金属敲击声（像是锤打铁器或施工的声音），声音响亮且持续，对说话人造成了明显的干扰。",
        "reasoning_content": "The user wants the audio transcribed and the background environment described.\n\n1.  **Transcription:** I hear a male voice speaking in Mandarin Chinese. He says, \"哎呀，这样我还怎么安静工作啊\" (Aiya, how can I work quietly like this?).\n2.  **Environment Sound:** There is a very distinct, rhythmic metallic clanking sound in the background. It sounds like someone is hammering metal or working with tools nearby. It's loud and repetitive.\n\nCombining these into the final response.",
        "role": "assistant"
      }
    }
  ],
  "created": 1792657756,
  "id": "chatcmpl-35486176-6b26-9b4d-968c-a5a599d5659f",
  "model": "qwen3.8-omni-flash",
  "object": "chat.completion",
  "usage": {
    "completion_tokens": 165,
    "completion_tokens_details": {
      "reasoning_tokens": 109,
      "text_tokens": 56
    },
    "prompt_tokens": 111,
    "prompt_tokens_details": {
      "audio_tokens": 44,
      "cached_tokens": 0,
      "text_tokens": 67
    },
    "total_tokens": 276
  }
}

🚗 音频理解URL（流式输出）调用示例 

# ============================================================================
#              qwen3.8-omni-flash 音频理解示例 (URL方式 - 流式输出)
# ============================================================================
# 功能说明：通过URL方式上传音频，使用流式输出实时获取模型响应
# 模型名称：qwen3.8-omni-flash
# ============================================================================

import requests
import json
import time

# ----------------------------------------------------------------------------
#                              API 配置信息
# ----------------------------------------------------------------------------
API_KEY = "sk-******************************************"  # API密钥
BASE_URL = "https://api.apexrouter.top/v1/chat/completions"            # 接口地址

# ----------------------------------------------------------------------------
#                              请求头配置
# ----------------------------------------------------------------------------
headers = {
    "Authorization": f"{API_KEY}",  # 认证信息
    "Content-Type": "application/json"     # 内容类型
}

# ----------------------------------------------------------------------------
#                              请求体构建
# ----------------------------------------------------------------------------
data = {
    # 指定使用的模型
    "model": "qwen3.8-omni-flash",

    # 消息列表
    "messages": [
        {
            "role": "user",
            "content": [
                {
                    # 文本提示词：指定音频处理任务（如转写为文字）
                    "type": "text",
                    "text": "请将这段音频转写为文字，并描述音频中的环境声音。"
                },
                {
                    "type": "input_audio",      # 音频输入类型
                    "input_audio": {
                        # 音频文件URL地址
                        "data": "https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20240916/xvappi/%E8%A3%85%E4%BF%AE%E5%99%AA%E9%9F%B3.wav",
                        # 音频格式（支持 wav、mp3 等）
                        "format": "wav"
                    }
                }
            ]
        }
    ],

    # ------------------------------------------------------------------------
    #                           流式输出配置
    # ------------------------------------------------------------------------
    "stream": True,                 # 启用流式输出
    "stream_options": {
        "include_usage": True       # 在流式响应中包含Token用量统计
    }
}


# ============================================================================
#                              主函数
# ============================================================================
def main():
    """
    主函数：发送流式请求并处理响应
    功能包括：实时输出、首Token延迟计算、Token用量统计
    """
    print("开始请求...")

    # ------------------------------------------------------------------------
    #                           初始化计时与变量
    # ------------------------------------------------------------------------
    start_time = time.time()      # 记录请求开始时间
    first_token_time = None       # 首个Token到达时间
    full_response = ""            # 完整响应内容
    usage_info = None             # Token用量信息

    # ------------------------------------------------------------------------
    #                           发送流式请求
    # ------------------------------------------------------------------------
    try:
        response = requests.post(
            BASE_URL,
            headers=headers,
            json=data,
            stream=True               # 启用流式接收
        )
        response.raise_for_status()   # 检查HTTP状态码

        # --------------------------------------------------------------------
        #                       逐行处理流式响应
        # --------------------------------------------------------------------
        for line in response.iter_lines():
            if line:
                line = line.decode('utf-8')

                # 处理SSE格式数据（以"data: "开头）
                if line.startswith("data: "):
                    json_str = line[6:]           # 去掉 "data: " 前缀

                    # 检测流结束标志
                    if json_str == "[DONE]":
                        break

                    try:
                        chunk = json.loads(json_str)

                        # 提取并输出内容片段
                        if "choices" in chunk and len(chunk["choices"]) > 0:
                            delta = chunk["choices"][0].get("delta", {})
                            content = delta.get("content", "")

                            if content:
                                # 记录首Token时间
                                if first_token_time is None:
                                    first_token_time = time.time()
                                # 实时输出内容（不换行）
                                print(content, end="", flush=True)
                                full_response += content

                        # 提取Token用量信息
                        if "usage" in chunk:
                            usage_info = chunk["usage"]

                    except json.JSONDecodeError:
                        continue

        print("\n")

    # ------------------------------------------------------------------------
    #                           异常处理
    # ------------------------------------------------------------------------
    except requests.exceptions.RequestException as e:
        print(f"\n请求错误: {e}")
        return

    # ========================================================================
    #                           统计信息输出
    # ========================================================================
    end_time = time.time()
    total_time = end_time - start_time

    print("=" * 50)
    print("统计信息:")
    print(f"  总耗时: {total_time:.2f}秒")

    # 首Token延迟 (TTFT: Time To First Token)
    if first_token_time:
        ttft = first_token_time - start_time
        print(f"  首Token延迟: {ttft:.2f}秒")

    # Token用量详情
    if usage_info:
        prompt_tokens = usage_info.get("prompt_tokens", 0)
        completion_tokens = usage_info.get("completion_tokens", 0)
        total_tokens = usage_info.get("total_tokens", 0)

        print(f"  输入Tokens: {prompt_tokens}")
        print(f"  输出Tokens: {completion_tokens}")
        print(f"  总Tokens: {total_tokens}")

        # 计算输出速度
        if total_time > 0 and completion_tokens > 0:
            tokens_per_sec = completion_tokens / total_time
            print(f"  输出速度: {tokens_per_sec:.2f} tokens/秒")


# ============================================================================
#                              程序入口
# ============================================================================
if __name__ == "__main__":
    main()

📥 返回示例（实测输出） 

开始请求...
**音频转写：**
哎呀，这样我还怎么安静工作啊

**环境声音描述：**
背景中有非常清晰且规律的金属敲击声（类似锤击或工具碰撞的声音），这种噪音很大且持续不断，显然是说话者抱怨无法安静工作的原因。

==================================================
统计信息:
  总耗时: 5.23秒
  首Token延迟: 4.17秒
  输入Tokens: 111
  输出Tokens: 179
  总Tokens: 290
  输出速度: 34.23 tokens/秒

🚀音频理解本地（base64） 

# ============================================================================
#              qwen3.8-omni-flash 音频理解示例 (本地文件 - Base64方式)
# ============================================================================
# 功能说明：读取本地音频文件，转换为Base64编码后发送给模型进行理解
# 模型名称：qwen3.8-omni-flash
# 适用场景：本地音频文件处理、无公网URL的音频分析
# ============================================================================

import base64
import requests

# ----------------------------------------------------------------------------
#                              API 配置信息
# ----------------------------------------------------------------------------
api_key = "sk-************************************************"  # API密钥
url = "https://api.apexrouter.top/v1/chat/completions"                 # 接口地址

# ----------------------------------------------------------------------------
#                              音频文件配置
# ----------------------------------------------------------------------------
# 填写本地音频文件路径，支持 wav、mp3 等格式
# 留空则使用默认示例数据
file_path = "qwen/装修噪音.wav"  # 例如: "C:/audio/test.mp3"


# ============================================================================
#                              工具函数
# ============================================================================
def file_to_base64(path):
    """
    读取本地文件并转换为Base64编码字符串

    参数:
        path: 文件路径
    返回:
        Base64编码的字符串
    """
    with open(path, "rb") as f:
        return base64.b64encode(f.read()).decode("utf-8")


# ----------------------------------------------------------------------------
#                           音频数据准备
# ----------------------------------------------------------------------------
# 根据是否填写文件路径，自动转换本地文件或使用示例数据
if file_path:
    audio_data = f"data:;base64,{file_to_base64(file_path)}"
else:
    # 默认示例数据（Base64编码的音频片段）
    audio_data = "data:;base64,SUQzBAAAAAAAI1RTU0UAAAAPAAADTGF2ZjU4LjI5...."

# ----------------------------------------------------------------------------
#                              请求头配置
# ----------------------------------------------------------------------------
headers = {
    "Authorization": f"{api_key}",  # 认证信息
    "Content-Type": "application/json"     # 内容类型
}

# ----------------------------------------------------------------------------
#                              请求体构建
# ----------------------------------------------------------------------------
data = {
    # 指定使用的模型
    "model": "qwen3.8-omni-flash",

    # 消息列表
    "messages": [
        {
            "role": "user",
            "content": [
                {
                    # 文本提示词：指定音频处理任务（如转写为文字）
                    "type": "text",
                    "text": "请将这段音频转写为文字，并描述音频中的环境声音。"
                },
                {
                    "type": "input_audio",      # 音频输入类型
                    "input_audio": {
                        "data": audio_data,     # Base64编码的音频数据
                        "format": "wav"         # 音频格式（支持 wav、mp3 等）
                    }
                }
            ]
        }
    ]
}

# ----------------------------------------------------------------------------
#                           发送请求并输出结果
# ----------------------------------------------------------------------------
response = requests.post(url, headers=headers, json=data)
print(response.json())

📥 返回示例 

{
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "message": {
        "content": "**转写文本：**\n哎呀，这样我还怎么安静工作啊\n\n**环境声音描述：**\n音频背景中充满了持续且响亮的金属敲击声（像是锤击金属的声音），节奏很快且重复，听起来像是附近的施工或打铁作业，严重干扰了说话者的工作环境。",
        "reasoning_content": "The user wants the audio transcribed and the environmental sounds described.\n\n1.  **Transcription:** I hear a male voice speaking in Mandarin Chinese. He says, \"哎呀，这样我还怎么安静工作啊\" (Aiya, zheyang wo hai zenme anjing gongzuo a). This translates to something like \"Oh come on, how am I supposed to work quietly like this?\"\n2.  **Environmental Sound:** Throughout the clip, there is a very distinct, rhythmic metallic clanking or hammering sound. It sounds like someone is hitting metal with a hammer, possibly construction work or blacksmithing nearby. The sound is loud and repetitive.\n\nCombining these observations into the final response.",
        "role": "assistant"
      }
    }
  ],
  "created": 1792657848,
  "id": "chatcmpl-5eaa2d7f-e48d-9a19-bc8b-d93e378e378d",
  "model": "qwen3.8-omni-flash",
  "object": "chat.completion",
  "usage": {
    "completion_tokens": 212,
    "completion_tokens_details": {
      "reasoning_tokens": 148,
      "text_tokens": 64
    },
    "prompt_tokens": 111,
    "prompt_tokens_details": {
      "audio_tokens": 44,
      "cached_tokens": 0,
      "text_tokens": 67
    },
    "total_tokens": 323
  }
}

© 2026 知能API qwen3.8-omni-flash 音频理解（STT）

 返回列表

与知能科技携手，共创数字化未来

欢迎沟通需求，我们将提供专业评估与建议。
了解更多