【AIGC】使用Python实现科大讯飞语音服务ASR转录功能：完整指南

参数名	说明
appId	讯飞开发者平台分配的应用ID
secret_key	用于生成签名的密钥
ts	时间戳，单位为秒
signa	认证签名，由appId、ts和secret_key计算得出

3.2 上传参数

参数名	说明
fileName	音频文件名称
fileSize	文件大小（若使用URL方式可随意填写）
duration	音频时长（单位秒，可随机填写）
language	语言（cn代表中文）
audioMode	上传模式（urlLink 代表通过URL上传）
audioUrl	音频文件的URL（需要URL编码）

3.3 查询结果参数

参数名	说明
orderId	订单ID，用于查询转写结果
resultType	返回结果类型（transfer 表示最终转写文本）

成功

{
 "code": "000000",
 "descInfo": "success",
 "content": {
  "orderId": "DKHJQ202209021522090215490FAAE7DD0008C",
  "taskEstimateTime": 28000
 }
}

失败

{
 "code": "26600",
 "descInfo": "转写业务通用错误"
}

3.4 orderResult 字段

参数名	类型	说明
lattice	List	做顺滑功能的识别结果
lattice2	List	未做顺滑功能的识别结果，当开启顺滑和后语规整后 orderResult 才返回 lattice2 字段（需要开通权限）
label	Object	转写结果标签信息，用于补充转写结果相关信息，标记转写结果角色和声道的对应关系

3.5 Lattice 字段

参数名	类型	说明
json_1best	String	单个 VAD 的结果的 JSON 内容

3.6 json_1best 字段

参数名	类型	说明
st	Object	单个句子的结果对象

3.7 st 字段

参数名	类型	说明
bg	String	单个句子的开始时间，单位毫秒
ed	String	单个句子的结束时间，单位毫秒
rl	String	分离的角色编号，取值正整数，需开启角色分离的功能才返回对应的分离角色编号
rt	List	输出词语识别结果集合

4. Python代码实现

4.1 生成签名

def generate_signa(self):
    """
    生成签名
    :return: 签名字符串
    """
    appid = self.appid
    ts = self.ts
    base_string = appid + ts
    
    md5 = hashlib.md5()
    md5.update(base_string.encode('utf-8'))
    base_string_md5 = md5.hexdigest()
    
    key = self.secret_key.encode('utf-8')
    message = base_string_md5.encode('utf-8')
    hmac_obj = hmac.new(key, message, hashlib.sha1)
    signa = base64.b64encode(hmac_obj.digest()).decode('utf-8')
    
    return signa

4.2 上传音频文件

1、文件上传
#概述
首先调用文件上传接口，上传待转写音频文件的基本信息（文件名、大小等）和相关的可配置参数。
调用成功，返回订单ID（orderId，用于查询结果或者联调排查问题时使用），是后续接口的必传参数。
#请求示例

https://raasr.xfyun.cn/v2/api/upload?duration=200&signa=Je5YsBvPcsbB4qy8Qvzd367fiv0%3D&fileName=%E9%98%B3%E5%85%89%E6%80%BB%E5%9C%A8%E9%A3%8E%E9%9B%A8%E5%90%8E.speex-wb&fileSize=11895&sysDicts=uncivilizedLanguage&appId=3e79d91c&ts=1662101767

#URL

 POST https: //raasr.xfyun.cn/v2/api/upload

#请求头

Content-Type: application/json; charset=UTF-8,Chunked: false

#signa生成

def upload(self):
    """
    上传音频文件
    :return: 上传响应结果
    """
    try:
        file_name = os.path.basename(urllib.parse.unquote(self.audio_url))
        param_dict = {
            "appId": self.appid,
            "signa": self.signa,
            "ts": self.ts,
            "fileName": file_name,
            "fileSize": "10000",
            "duration": "200",
            "language": "cn",
            "audioMode": "urlLink",
            "audioUrl": urllib.parse.quote(self.audio_url, safe='')
        }
        url = lfasr_host + api_upload + "?" + urllib.parse.urlencode(param_dict)
        response = requests.post(url, headers={"Content-Type": "application/json; charset=UTF-8"})
        result = json.loads(response.text)
        
        if str(result.get('code', '')) != '000000':
            return None
        
        return result
    except Exception as e:
        return None

4.3 获取转写结果

status == 4

def get_result(self):
    """
    获取转写结果
    :return: 转写结果
    """
    uploadresp = self.upload()
    if not uploadresp:
        return None
    
    try:
        orderId = uploadresp['content']['orderId']
        param_dict = {
            "appId": self.appid,
            "signa": self.signa,
            "ts": self.ts,
            "orderId": orderId,
            'roleType':1,
            'roleNum':2,
            "resultType": "transfer"
        }
        
        url = lfasr_host + api_get_result + "?" + urllib.parse.urlencode(param_dict)
        status = 3
        while status == 3:
            response = requests.post(url, headers={"Content-Type": "application/json; charset=UTF-8"})
            result = json.loads(response.text)
            
            if str(result.get('code', '')) != '000000':
                continue
                
            status = result['content']['orderInfo']['status']
            if status == 4:
                return result
            time.sleep(5)
        
        return None
    except Exception as e:
        return None

查询结构返回数据

{
 "code": "000000",
 "descInfo": "success",
 "content": {
  "orderInfo": {
   "orderId": "DKHJQ2022090510220905100562536FEF00062",
   "failType": 0,
   "status": 4,
   "originalDuration": 200,
   "realDuration": 1878
  },
  "orderResult": "{"lattice":[{"json_1best":"{"st":{"sc":"0.86","pa":"0","rt":[{"ws":[{"cw":[{"w":"这","wp":"n","wc":"1.0000"}],"wb":1,"we":16},{"cw":[{"w":"是","wp":"n","wc":"1.0000"}],"wb":17,"we":36},{"cw":[{"w":"一","wp":"n","wc":"1.0000"}],"wb":37,"we":52},{"cw":[{"w":"条","wp":"n","wc":"1.0000"}],"wb":53,"we":80},{"cw":[{"w":"测试","wp":"n","wc":"1.0000"}],"wb":81,"we":116},{"cw":[{"w":"音频","wp":"n","wc":"1.0000"}],"wb":117,"we":172},{"cw":[{"w":"。","wp":"p","wc":"0.0000"}],"wb":172,"we":172},{"cw":[{"w":"","wp":"g","wc":"0.0000"}],"wb":172,"we":172}]}],"bg":"50","rl":"0","ed":"1840"}}"}],"lattice2":[{"lid":"0","end":"1840","begin":"50","json_1best":{"st":{"sc":"0.86","pa":"0","rt":[{"nb":"1","nc":"1.0","ws":[{"cw":[{"w":"这","wp":"n","wc":"1.0000"}],"wb":1,"we":16},{"cw":[{"w":"是","wp":"n","wc":"1.0000"}],"wb":17,"we":36},{"cw":[{"w":"一","wp":"n","wc":"1.0000"}],"wb":37,"we":52},{"cw":[{"w":"条","wp":"n","wc":"1.0000"}],"wb":53,"we":80},{"cw":[{"w":"测试","wp":"n","wc":"1.0000"}],"wb":81,"we":116},{"cw":[{"w":"音频","wp":"n","wc":"1.0000"}],"wb":117,"we":172},{"cw":[{"w":"。","wp":"p","wc":"0.0000"}],"wb":172,"we":172},{"cw":[{"w":"","wp":"g","wc":"0.0000"}],"wb":172,"we":172}]}],"pt":"reserved","bg":"50","si":"0","rl":"0","ed":"1840"}},"spk":"段落-0"}]}",
  "taskEstimateTime": 0
 }

4.4 解析转写结果

def parse_result(self, result_json):
    """
    解析转写结果，按说话人分组
    """
    try:
        result = json.loads(result_json)
        speakers = {}
        if 'lattice2' in result:
            for item in result['lattice2']:
                speaker = item.get('spk', '未知')
                json_1best = json.loads(item['json_1best'])
                text = "".join(cw['w'] for rt in json_1best.get('st', {}).get('rt', []) for ws in rt.get('ws', []) for cw in ws.get('cw', []) if 'w' in cw)
                speakers.setdefault(speaker, []).append(text)
        return speakers
    except Exception as e:
        return None

5. 请求与返回示例

5.1 成功返回示例

{
 "code": "000000",
 "descInfo": "success",
 "content": {
  "orderId": "DKHJQ202209021522090215490FAAE7DD0008C",
  "taskEstimateTime": 28000
 }
}

5.4 异步回调

1. 转写结束异步回调状态

当订单转写流程结束时会回调用户（如果录音文件转写接口 upload 传了callbackUrl），会把订单号和订单状态返回，具体的格式和参数说明如下：回调地址示例：

GET http://ip:prot/server/xxx?orderId=DKHJQ202004291620042916580FBC96690001F&status=1

注意 QPS 和 API 限流

CODE: 100012 26603

可进行队列延迟重试处理

6. 文档地址

讯飞ASR文档

7. 结论

本文详细讲解了讯飞ASR API的使用流程，包括如何生成签名、上传音频、查询结果并解析返回数据。希望这篇文章对你有所帮助！如果对你有帮助，帮忙给个一键三连，求求了，各位吴彦祖，刘亦菲们

文章来源于互联网:【AIGC】使用Python实现科大讯飞语音服务ASR转录功能：完整指南

相关推荐: cursor 和 GitHub copilot 最强竞品：开源 Void 编辑器会取代他们吗？

你是否在使用 GitHub Copilot 或 Cursor 的时候，觉得它们的智能补全虽然强大，但总有些地方不尽如人意？比如，价格高昂，或者一些功能限制让人抓狂？如果你有类似的痛点，那么今天我要给你介绍一款新工具——Void，这款编辑器号称是开源的 Curs…

【AIGC】使用Python实现科大讯飞语音服务ASR转录功能：完整指南

讯飞ASR转写API完整指南

1. 引言

2. 讯飞ASR API介绍

3. API参数说明

3.1 认证参数