def vad():
logger.info(vad_sh)
os.system(vad_sh)
if not os.path.exists(segment):
logger.error("vad failed！")
os._exit(0)
else:
logger.info("vad succsee！")
def run(self):
"""
"""
logger.debug("遍历wav_name,text信息")
wav_suf = "wav"
counter = 0
whole_wavs = set()
wavs_info_map = {}
logger.info(f"按照 【%s】 进行分包!" %
self.custom_classfy if self.custom_classfy else "SPK")
for spk, wav_name, content in self.mklines(): # 遍历说话人id，音频名，音频内容 信息
wav_time = self.wav_time_map.get(wav_name)
if wav_time is None:
logger.warning("未获取到音频时间 [%s]" % wav_name)
continue
wav_info = [ # 填充新平台文本格式
{
"Wav_name":
wav_name,
"Length_time":
wav_time,
"Data": [{
"text": content,
"start_time": 0,
"end_time": wav_time
}],
"Wav_suf":
wav_suf,
}
]
whole_wavs.add(wav_name)
if self.custom_classfy: # 指定分包数的模式
ids = counter
if ids not in wavs_info_map:
wavs_info_map[ids] = [wav_info]
else:
if len(wavs_info_map[ids]) == self.custom_classfy - 1:
counter += 1
wavs_info_map[ids].append(wav_info)
else: # 默认分包模式
if spk not in wavs_info_map:
wavs_info_map[spk] = [wav_info]
else:
wavs_info_map[spk].append(wav_info)
whole_sum = len(whole_wavs)
result = (wavs_info_map, whole_sum)
return result
def update_graph():
# 中文分词
cmd = f"python3 {word_seg_py}"
logger.info(cmd)
os.system(cmd)
# 更新模型
logger.info(update_graph_sh)
os.system(update_graph_sh)
def main(wavspath, outputpath, value, tocut):
result_error = {}
for root, dirs, files in os.walk(wavspath):
for file in files:
if not file.endswith('wav'):
continue
package_name = file.rstrip('.wav')
logger.info(f'处理{package_name}中.')
wav_path = os.path.join(root, package_name + '.wav')
txt_path = os.path.join(root, package_name + '.txt')
wavmaps = read_txttime(txt_path)
wavslist = sorttime(wavmaps, txt_path)
tail_time = read_tail_time(wavslist)
wave_time = read_wavtime(wav_path)
time_ = tail_time - wave_time
diff_time = round(time_, 3)
logger.info(f'时间点差为{diff_time}.')
if not judgement_time(time_, threshold=value):
result_error[package_name] = {'tail_time': tail_time, 'wave_time': wave_time}
continue
if tocut:
outputpath = mkdir(os.path.join(outputpath, package_name))
pool = Pool(processes=8)
logger.info(f'切割 {file} 中...')
for item in wavslist:
wavname = item.get('wavname')
outputwav = os.path.join(outputpath, wavname)
start = item.get('start')
end = item.get('end')
pool.apply_async(wavcut, (wav_path, outputwav, start, end))
pool.close()
pool.join()
logger.info('检测完成！')
if result_error:
logger.error('%s 未通过检测，需手动调整切割 [%s]' % (list(result_error.keys()), result_error))
else:
logger.info('全部音频通过检测！')
def sorttime(maps, txt_path):
wavinfo = list()
with open(txt_path, 'w', encoding='utf-8') as fw:
fw.write(maps.pop('header'))
for wavname, values in sorted(maps.items(), key=lambda x: float(x[1].get('start'))):
start = values.get('start')
end = values.get('end')
text = values.get('text')
fw.write(f"{wavname}\t[{start},{end}]\t{text}\n")
wavinfo.append({'wavname': wavname, 'start': float(start), 'end': float(end)})
logger.info('时间排序完成！')
return wavinfo
def proc(file_path, rate, channel, bit, target_dir, func):
file_name = os.path.basename(file_path)
target_path = os.path.join(target_dir, file_name)
if func == "addhead":
cmd = "sox -t raw -c %s -e signed-integer -b %s -r %s %s %s" % (
channel,
bit,
rate,
file_path,
target_path,
)
else:
cmd = f"sox {file_path} -r {rate} -b {bit} -c {channel} {target_path}"
logger.info(cmd)
os.system(cmd)
def run_decode():
# 数据vad
segment_dir = f"output/{project_name}_seg"
if not os.path.exists(os.path.join(segment_dir, "segments")):
vad()
else:
logger.info("sgements file has exist, vad pass!")
# 数据准备
cmd = f"cp -r {segment_dir} data/"
logger.info(cmd)
os.system(cmd)
# 解码操作
logger.info(decode_sh)
os.system(decode_sh)
logger.info("####------Done!--------####")
def run_decode():
# 数据vad
if not os.path.exists(segment):
vad()
else:
logger.info("sgements file has exist, vad pass!")
# 数据准备
cmd = f"cp -r {segment_dir} data/"
logger.info(cmd)
os.system(cmd)
# 解码操作
logger.info(decode_sh)
os.system(decode_sh)
if os.path.exists(decode_file):
logger.info("####------解码完成!--------####")
else:
logger.error("####------解码失败!--------####")
def vad():
logger.info(vad_sh)
os.system(vad_sh)
else:
logger.info("sgements file has exist, vad pass!")
# 数据准备
cmd = f"cp -r {segment_dir} data/"
logger.info(cmd)
os.system(cmd)
# 解码操作
logger.info(decode_sh)
os.system(decode_sh)
logger.info("####------Done!--------####")
if __name__ == "__main__":
if func == "update":
update_graph()
os._exit(0)
elif func == "decode":
run_decode()
os._exit(0)
elif func == "vad":
vad()
os._exit(0)
else:
update_graph()
run_decode()
logger.info("DONE!")
if __name__ == "__main__":
file_exists(project_wavs_path)
file_exists(project_txts_path)
if not check_valid(project_txts_path):
sys.exit()
if not os.path.exists(time_map):
dic_map = wavtime_map(project_wavs_path)
with open(time_map, "w", encoding="utf-8") as f:
json.dump(dic_map, f)
logger.info("保存timemap成功！ 共%s条" % len(dic_map.keys()))
if not os.path.exists(time_map):
logger.info("time_map 未生成!!!")
sys.exit()
with open(time_map, "r", encoding="utf-8") as f:
dic_map = json.loads(f.read())
logger.info("载入音频map完成！ \n--> %s" %
{key: dic_map[key]
for key in list(dic_map.keys())[:6]})
result_map, whsum = CreatPTFile(project_txts_path, dic_map, classify).run()
save_to_file(save_upfile, result_map)