周五在公司内部做了一个分享,手机全程录制的音频,想着整理成文字稿,但整个分享会时间 50 分钟,内容太多,就想着剥削下电脑吧,再次感慨下,懒人真是适合当程序员。

先在网上找找有没有现成的脚本可用,逛了一圈发现没有太好的,还是要亲力亲为。百度 API 和讯飞 API 是相对成熟的,开始一直鼓捣百度语音 API 接口,但莫名的 bug 一直卡着不动,就改为使用讯飞 API 了。这个开发文档写的比较完善,Demo 脚本也详细,研究了两个小时,就把我 50min 的讲座内容存到 txt 文件中了。本脚本直接使用的是讯飞提供的完整 Python 脚本,只是这个脚本不能将转录的文字存储到文件中,故结尾加了一个文件写入函数而已。

操作步骤:

  • 在讯飞开放平台注册账号,点击「语音识别」,新建一个应用,再选择「语音转写」,获取密钥;
  • 查看音频采样率,我用苹果手机录制的音频采样率为 48k,需要降到 16k 才能处理,音频属性需要 16k 或 8k;
  • 因需要使用 Requests 请求接口,故需要联网运行;
  • 更改 RequestApi 的三个输入参数,运行脚本;
  • 脚本运行完毕后,会生成一个 txt 文本。转录好的文字,就存在这个文件中。

需要注意的是,讯飞提供的免费转录时间只有几个小时,之后就要收费了,所以谨慎使用。转录出来的效果差强人意,识别的不是很精准,错误较多,不过好在不用自己一个字一个字的敲了,还是能凑合用一下的,如果收费的话,那感觉真心不值。


# -*- coding: utf-8 -*- 

import base64
import hashlib
import hmac
import json
import os
import time
import requests

lfasr_host = 'http://raasr.xfyun.cn/api'

# 请求的接口名
api_prepare = '/prepare'
api_upload = '/upload'
api_merge = '/merge'
api_get_progress = '/getProgress'
api_get_result = '/getResult'
# 文件分片大小10M
file_piece_sice = 10485760

class SliceIdGenerator:
    """slice id生成器"""

    def __init__(self):
        self.__ch = 'aaaaaaaaa`'

    def getNextSliceId(self):
        ch = self.__ch
        j = len(ch) - 1
        while j >= 0:
            cj = ch[j]
            if cj != 'z':
                ch = ch[:j] + chr(ord(cj) + 1) + ch[j + 1:]
                break
            else:
                ch = ch[:j] + 'a' + ch[j + 1:]
                j = j - 1
        self.__ch = ch
        return self.__ch


class RequestApi(object):
    def __init__(self, appid, secret_key, upload_file_path):
        self.appid = appid
        self.secret_key = secret_key
        self.upload_file_path = upload_file_path

    def gene_params(self, apiname, taskid=None, slice_id=None):
        appid = self.appid
        secret_key = self.secret_key
        upload_file_path = self.upload_file_path
        ts = str(int(time.time()))
        m2 = hashlib.md5()
        m2.update((appid + ts).encode('utf-8'))
        md5 = m2.hexdigest()
        md5 = bytes(md5, encoding='utf-8')
        # 以secret_key为key, 上面的md5为msg, 使用hashlib.sha1加密结果为signa
        signa = hmac.new(secret_key.encode('utf-8'), md5, hashlib.sha1).digest()
        signa = base64.b64encode(signa)
        signa = str(signa, 'utf-8')
        file_len = os.path.getsize(upload_file_path)
        file_name = os.path.basename(upload_file_path)
        param_dict = {}

        if apiname == api_prepare:
            # slice_num是指分片数量,如果您使用的音频都是较短音频也可以不分片,直接将slice_num指定为1即可
            slice_num = int(file_len / file_piece_sice) + (0 if (file_len % file_piece_sice == 0) else 1)
            param_dict['app_id'] = appid
            param_dict['signa'] = signa
            param_dict['ts'] = ts
            param_dict['file_len'] = str(file_len)
            param_dict['file_name'] = file_name
            param_dict['slice_num'] = str(slice_num)
        elif apiname == api_upload:
            param_dict['app_id'] = appid
            param_dict['signa'] = signa
            param_dict['ts'] = ts
            param_dict['task_id'] = taskid
            param_dict['slice_id'] = slice_id
        elif apiname == api_merge:
            param_dict['app_id'] = appid
            param_dict['signa'] = signa
            param_dict['ts'] = ts
            param_dict['task_id'] = taskid
            param_dict['file_name'] = file_name
        elif apiname == api_get_progress or apiname == api_get_result:
            param_dict['app_id'] = appid
            param_dict['signa'] = signa
            param_dict['ts'] = ts
            param_dict['task_id'] = taskid
        return param_dict

    def gene_request(self, apiname, data, files=None, headers=None):
        response = requests.post(lfasr_host + apiname, data=data, files=files, headers=headers)
        result = json.loads(response.text)
        if result["ok"] == 0:
            print("{} success:".format(apiname) + str(result))
            return result
        else:
            print("{} error:".format(apiname) + str(result))
            exit(0)

    # 预处理
    def prepare_request(self):
        return self.gene_request(apiname=api_prepare,
                                 data=self.gene_params(api_prepare))

    # 上传
    def upload_request(self, taskid, upload_file_path):
        file_object = open(upload_file_path, 'rb')
        try:
            index = 1
            sig = SliceIdGenerator()
            while True:
                content = file_object.read(file_piece_sice)
                if not content or len(content) == 0:
                    break
                files = {
                    "filename": self.gene_params(api_upload).get("slice_id"),
                    "content": content
                }
                response = self.gene_request(api_upload,
                                             data=self.gene_params(api_upload, taskid=taskid,
                                                                   slice_id=sig.getNextSliceId()),
                                             files=files)
                if response.get('ok') != 0:
                    # 上传分片失败
                    print('upload slice fail, response: ' + str(response))
                    return False
                print('upload slice ' + str(index) + ' success')
                index += 1
        finally:
            'file index:' + str(file_object.tell())
            file_object.close()
        return True

    # 合并
    def merge_request(self, taskid):
        return self.gene_request(api_merge, data=self.gene_params(api_merge, taskid=taskid))

    # 获取进度
    def get_progress_request(self, taskid):
        return self.gene_request(api_get_progress, data=self.gene_params(api_get_progress, taskid=taskid))

    # 获取结果
    def get_result_request(self, taskid):
        return self.gene_request(api_get_result, data=self.gene_params(api_get_result, taskid=taskid))

    def all_api_request(self):
        # 1. 预处理
        pre_result = self.prepare_request()
        taskid = pre_result["data"]
        # 2 . 分片上传
        self.upload_request(taskid=taskid, upload_file_path=self.upload_file_path)
        # 3 . 文件合并
        self.merge_request(taskid=taskid)
        # 4 . 获取任务进度
        while True:
            # 每隔20秒获取一次任务进度
            progress = self.get_progress_request(taskid)
            progress_dic = progress
            if progress_dic['err_no'] != 0 and progress_dic['err_no'] != 26605:
                print('task error: ' + progress_dic['failed'])
                return
            else:
                data = progress_dic['data']
                task_status = json.loads(data)
                if task_status['status'] == 9:
                    print('task ' + taskid + ' finished')
                    break
                print('The task ' + taskid + ' is in processing, task status: ' + str(data))

            # 每次获取进度间隔20S
            time.sleep(20)
        # 5 . 获取结果
        return self.get_result_request(taskid=taskid)


# requests版本可能要在2.20.0以上
# 输入讯飞开放平台的appid,secret_key和待转写的文件路径
if __name__ == '__main__':
    api = RequestApi(appid="f3b985fe", secret_key="d46271748fc66b0fd29c7a561fb9371d", upload_file_path="voice6.wav")
    txtResult = api.all_api_request()

    fileToSave = open('speech2Word.txt', 'w')
    listData = eval(txtResult['data'])
    for ii in range(len(listData)):
        print(listData[ii]['onebest'])
        fileToSave.write(listData[ii]['onebest'])
    # print(dictText['data'][1]['onebest'])
    fileToSave.close()

参考文档:讯飞 API_语音听写