合作者：@Yasar @NanoApe
附件内是一个 pcap 文件，里面有很多个 UDP，Payload 格式经过分析，大致如下：
@Yasar 发现 Payload 是音频数据的浮点数格式，把每个 UDP 的 payload 提取出来，转成 wav 文件：
import numpy as np
import wave
# get data from pcap omitted
arr_u8 = np.frombuffer(data, dtype=np.uint8)
MU_LAW_BIAS = 0x84
def mulaw_decode_byte(mu):
mu = ~mu & 0xFF
sign = mu & 0x80
exponent = (mu >> 4) & 0x07
mantissa = mu & 0x0F
magnitude = ((mantissa << 1) + 1) << (exponent + 2)
magnitude = magnitude - MU_LAW_BIAS
sample = -magnitude if sign else magnitude
return np.int16(max(-32768, min(32767, sample << 2)))
arr_mulaw_decoded = np.array([mulaw_decode_byte(int(b)) for b in arr_u8], dtype=np.int16)
with wave.open('ulaw_decoded_8k_s16.wav', 'wb') as wf:
wf.setnchannels(1)
wf.setsampwidth(2)
wf.setframerate(8000)
wf.writeframes(arr_mulaw_decoded.tobytes())
如此可以得到 1000+ 个 wav 文件。由于内容太多，所以丢给 ASR 进行语音识别：
# based on https://github.com/jiegec/video2srt
import argparse
import os
import base64
from re import S
from time import sleep
from tencentcloud.common import credential
from tencentcloud.common.exception.tencent_cloud_sdk_exception import TencentCloudSDKException
from tencentcloud.asr.v20190614 import asr_client, models
parser = argparse.ArgumentParser(
description='Recognize audio')
parser.add_argument('audios', metavar='AUDIO', type=str, nargs='+',
help='path to audio files')
parser.add_argument('--secret-id', type=str,
help='secret id')
parser.add_argument('--secret-key', type=str,
help='secret key')
args = parser.parse_args()
for audio in args.audios:
print('Processing {}'.format(audio))
base, ext = os.path.splitext(audio)
srt = '{}.srt'.format(base)
json = '{}.json'.format(base)
# Step 1: upload to tencent cloud
print('Upload to tencent cloud')
cred = credential.Credential(args.secret_id, args.secret_key)
client = asr_client.AsrClient(cred, "ap-shanghai")
data = open(audio, 'rb').read()
req = models.CreateRecTaskRequest()
req.EngineModelType = "16k_zh"
req.ChannelNum = 1
req.ResTextFormat = 2
req.SourceType = 1
req.ConvertNumMode = 3
req.FilterModal = 1
req.Data = base64.b64encode(data).decode('utf-8')
resp = client.CreateRecTask(req)
task_id = resp.Data.TaskId
# Step 2: retrieve result
while True:
print('Wait for result')
req = models.DescribeTaskStatusRequest()
req.TaskId = task_id
resp = client.DescribeTaskStatus(req)
if resp.Data.StatusStr == "success":
print(f'Save json result to {json}')
open(json, 'w').write(resp.to_json_string())
break
sleep(5)
# Step 3: save to srt
print(f'Save subtitle to {srt}')
result = resp.Data.Result
counter = 1
with open(srt, 'w') as f:
for line in result.split('\n'):
if len(line) == 0:
break
parts = line.split(']')
times = parts[0].split(',')
time_from = times[0][1:]
time_to = times[1][:-1]
print(counter, file=f)
print('0:{} --> 0:{}'.format(time_from, time_to), file=f)
print(']'.join(parts[1:]).strip(), file=f)
print('', file=f)
counter += 1
在结果里，找到一段强网杯 2025 相关的字符串：
==> ulaw_decoded_8k_s16_40259.srt <==
1
0:0:0.000 --> 0:0:5.14
第9届全网杯2025震撼来袭，你准备好了吗？
2
0:0:19.800 --> 0:0:44.70
四六我不您解后一条命令，6514663451427161661421466040145664160141451426071146466014214371606514214470。
啥玩意儿
@NanoApe 指出，这是一系列的 ascii 码，转成 8 进制后拼接的结果：
5 -> 53 (dec) -> 65 (oct)f -> 102 (dec) -> 146 (oct)因此对应的字符串是 5f3eb916bf08e610aeb09f60bc955bd8。可以得到一段 mp3，经过语音识别：
表兄近日可好，上回托您带的甘四担秋茶，家母嘱咐务必在辰时正过三刻前送到，切记用金丝锦盒装妥，此处潮气重，莫让干货瘦了霉，若赶得及时可赶得菊花开前便可让铺子开张，
一切安好，我会按照要求准备好抽查我该送到何地，
送至双鲤湖西岸，南山茶铺放右边第二个橱柜莫放错，
我已知悉你在那边可还安好，
一切安好，希望你我二人早日相见，
指日可待。茶叶送到了，但是晚了时日，茶铺看来只能另寻良辰吉日了，你在那边千莫保重
@NanoApe 根据 双鲤湖西岸南山茶铺，查到 1949 年 10 月金门战役的历史事件，结合已有信息：
最终得到：1949年10月24日8时45分于双鲤湖西岸南山茶铺。