# coding:utf-8

# 百度智能云语音批量识别m4a转txt（也可扩展其他音频格式，增加一两句代码即可）
# 作者：雪山凌狐
# 版本：1.1
# 更新时间：2020-02-15
# 网址：http://www.xueshanlinghu.com

# 使用说明请见README.md文件

import subprocess
import json
import os
import time
from decimal import Decimal

from aip import AipSpeech

# 配置部分=========================================
# 将你要识别的m4a文件（或其他你测试后支持的音频格式）放到本程序目录下的设定文件夹下，待识别。你也可以自己修改封装更通用的版本（我这里是为了简便）
asr_folder_name = "audio"
# 结束后是否删除pcm识别文件？推荐删除，否则可能会很多，下次可以再生成嘛
DELETE_PCM = True
""" 你的 APPID AK SK """
APP_ID = '你的 App ID'
API_KEY = '你的 Api Key'
SECRET_KEY = '你的 Secret Key'
# QPS（每秒请求数）限制，免费版目前是5。如果你是无限的QPS，可以把这个值调得非常大
QPS = 5
# 配置部分=========================================


client = AipSpeech(APP_ID, API_KEY, SECRET_KEY)

def get_piece(duration, piece_len=59):
	"""获取总的切片长度，每片的长度为piece_len"""
	pieces = duration // piece_len + 1
	return int(round(pieces, 0))

# 读取文件
def get_file_content(filePath):
	"""读取文件，通过二进制的方式读入"""
	with open(filePath, 'rb') as fp:
		return fp.read()

# 如果你测试能够支持转换mp3，wav，amr等格式，可以在下面的定义中添加即可（我没测试过，不确定），如ext=["m4a", "mp3", "wav", "amr"]
def get_filelist(folder_name, ext=["m4a"]):
	"""遍历限定的文件夹下的所有文件完整路径"""
	if not(folder_name.endswith("\\")):
		folder_name += "\\"
	final_filelist = []
	filelist = os.listdir(folder_name)
	# print(filelist)
	for i in filelist:
		path = folder_name + i
		if os.path.isfile(path):
			if os.path.splitext(i)[1].replace(".", "") in ext:
				final_filelist.append(path)
		elif os.path.isdir(path):
			# 进行子目录遍历，递归
			child_filelist = get_filelist(path, ext)
			final_filelist += child_filelist
	return final_filelist

def get_process(process_num, total_num, percent_sign="%"):
	"""获取百分比进度"""
	def remove_exponent(num, ndigits=2):
		"""如果数字结尾有多余的0，把多余的0去掉，方法摘自网络，做了部分修改改进"""
		num = Decimal(num)
		return round(num.to_integral(), ndigits) if num == num.to_integral() else round(num.normalize(), ndigits)

	process = round(process_num / total_num * 100, 2)
	process = remove_exponent(process)
	return str(process) + percent_sign

def get_process_sign(process_num, total_num, sign="="):
	"""获取百分比进度对应的进度条"""
	process = get_process(process_num, total_num, percent_sign="")
	process = int(round(float(process), 0))
	process = process // 2
	return sign * process + ">"

if __name__ == "__main__":
	# 获取项目开始时间
	start_time = time.time()

	project_path = os.path.dirname(os.path.abspath(__name__)) + "\\" + asr_folder_name
	# print(project_path)
	filenames = get_filelist(project_path)
	# print(filenames)

	file_len = len(filenames) # 有几个文件要识别的
	finished_count = 0 # 完成了几个文件初始化
	QPS_count = 0 # 目前的请求数初始化
	QPS_start_time = time.time() # 目前开始请求的时间，初始化
	for cur_file in filenames:
		# 获取文件路径、文件名等信息
		cur_file_folder = os.path.dirname(cur_file) + "\\"
		cur_file_name = os.path.basename(cur_file)
		cur_file_name_withoutext = os.path.splitext(cur_file_name)[0]
		# print(cur_file_folder, cur_file_name, cur_file_name_withoutext)

		# 获取音频时长信息duration
		cmd = 'ffprobe -v quiet -print_format json -show_streams "%s"' % cur_file
		sub = subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE,
			universal_newlines=True)
		# print(cmd)
		out, err = sub.communicate()
		sub.kill()
		del sub
		# print(out)
		out = json.loads(out)
		duration = out.get("streams")[0].get("duration")
		print(cur_file, "持续时间为：", duration, "秒")
		# 获取切片数
		pieces = get_piece(float(duration))
		print("切片数：", pieces)

		# 进行PCM格式转换。每59秒切片一次，因为目前百度智能云的语音识别最长支持单语音文件是60秒
		piece_len = 59
		result_name = cur_file_folder + cur_file_name_withoutext + ".txt"
		if os.path.exists(result_name):
			os.remove(result_name)
		with open(result_name, "a", encoding="utf-8") as f:
			for i in range(pieces):
				# 转码成pcm格式的文件，转换成百度智能云的语音识别能识别的pcm格式编码率等
				# 目前百度智能云语音识别要求的pcm格式的参数为：原始 PCM 的录音参数必须符合 16k 采样率、16bit 位深、单声道
				pcm_filename = cur_file_folder + cur_file_name_withoutext + ("_" + str(i + 1) if i > 0 else "") + ".pcm"
				cmd = 'ffmpeg -y -ss %s -t %s -i "%s" -avoid_negative_ts 1 -acodec pcm_s16le -f s16le -ac 1 -ar 16000 "%s"' \
						% (i * 59, piece_len, cur_file, pcm_filename)
				# print(cmd)

				sub = subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
				out, err = sub.communicate()
				sub.kill()
				del sub

				# 进行云端识别调用百度api
				print("云端识别[" + pcm_filename + "]这个文件中...")
				res = client.asr(get_file_content(pcm_filename), 'pcm', 16000)

				if res.get("err_no") == 0:
					result = res.get("result")
					for cur_line in result:
						print("[识别结果]", cur_line + "\n")
						f.write(cur_line + "\n")
				else:
					print(pcm_filename, "识别报错！")
					print(res)

				# 检查机制，避免每秒请求数过大
				QPS_count += 1
				if QPS_count >= QPS:
					QPS_end_time = time.time()
					if QPS_end_time - QPS_start_time < 1:
						print("休息一秒，QPS（每秒请求数）有限制，目前设定为：" + str(QPS) + "，避免频繁调用报错..." + "\n")
						# 实际休息1.1秒，更保险
						time.sleep(1.1)
					# 重新新的一轮统计
					QPS_start_time = time.time()
					QPS_count = 0

		print("[" + result_name, "]文件写入结束！" + "\n")
		finished_count += 1
		print("*" * 40)
		print("当前已完成进度为：" + get_process_sign(finished_count, file_len) + get_process(finished_count, file_len))
		print("*" * 40 + "\n")

	# 处理PCM中间文件
	if DELETE_PCM:
		print("正在删除pcm中间文件...")
		filenames = get_filelist(project_path, ext=["pcm"])
		for file in filenames:
			os.remove(file)
		print("删除完毕！" + "\n")

	end_time = time.time()
	print("总共%s个文件全部识别完毕！用时%s秒，感谢您的使用！" % (file_len, round(end_time - start_time, 2)))