---

0、一个fs的web配置界面预览

http://myfs.f3322.net:8020/
用户名：admin，密码：admin

FreeSWITCH界面安装参考：https://blog.csdn.net/jia198810/article/details/137820796

1、docker地址

registry.cn-hangzhou.aliyuncs.com/pbx/asr-47m            latest    616 MB
registry.cn-hangzhou.aliyuncs.com/pbx/asr-14m            latest    588 MB
registry.cn-hangzhou.aliyuncs.com/pbx/asr-124m           latest    653 MB
registry.cn-hangzhou.aliyuncs.com/pbx/my-funasr          latest    6.5G

2、使用

2.1 下载

# 下载对应的asr地址
podman pull registry.cn-hangzhou.aliyuncs.com/pbx/asr-124m

2.2 运行

# asr-124m  中英文 124M
# sherpa-ncnn-streaming-zipformer-bilingual-zh-en-2023-02-13
# 启动 -v为热词，可选，-e为变量，NUM_THREADS为最大线程，可选，下面的都一样
podman run -itd -p 8000:8000/tcp -v /root/hotwords.txt:/root/hotwords.txt -e NUM_THREADS=4 --name asr-124m asr-124m
# 查看日志
podman exec -it asr-124m tail -f /var/log/asr.log # asr-14m  中文 14M
# sherpa-ncnn-streaming-zipformer-zh-14m-2023-02-23-chinese
podman run -itd -p 8000:8000/tcp -v /root/hotwords.txt:/root/hotwords.txt -e NUM_THREADS=4 --name asr-14m asr-14m# asr-47m  中英文 47M
# sherpa-ncnn-streaming-zipformer-small-bilingual-zh-en-2023-02-16-bilingual-chinese-english
podman run -itd -p 8000:8000/tcp -v /root/hotwords.txt:/root/hotwords.txt -e NUM_THREADS=4 --name asr-47m asr-47m# Funasr  启动较慢
# 可以进入容器netstat -tnlp是否启动8000和10095 ,
#`offline`表示推理模式为一句话识别；
#`online`表示推理模式为实时语音识别;
#`2pass`表示为实时语音识别，并且说话句尾采用离线模型进行纠错；
# 用online吧，online效果还是可以的
podman run -itd -p 8000:8000 -p 10095:10095  -e MODE='online'  -v /root/hotwords.txt:/workspace/models/hotwords.txt --name my-funasr my-funasr

3、例子

3.1 下载

root@xiaojia-ts /h/xiaojia# podman pull registry.cn-hangzhou.aliyuncs.com/pbx/asr-124m                                                                                           (base) 
Trying to pull registry.cn-hangzhou.aliyuncs.com/pbx/asr-124m:latest...
Getting image source signatures
Copying blob fcbb30fb83d0 done   | 
Copying blob 5b4de9788e44 skipped: already exists  
Copying blob f686e60318db done   | 
Copying blob 095f960760ee skipped: already exists  
Copying blob 6721db44a395 done   | 
Copying config 8e4c94354d done   | 
Writing manifest to image destination
8e4c94354db2df8a0dea58c552e616afcc660acad95bd940f280c193129ab908

3.2 启动

root@xiaojia-ts /h/xiaojia# podman run -itd -p 8000:8000/tcp  -e NUM_THREADS=4 --name asr-124m asr-124m                                                                          (base) 
96878427c8d4e1b0b2bf4fe96386c253d10f95eac12fa32af4066747b4d2d231

3.3 编译mod_audio_fork或者mod_audio_stream模块使用

参考下面的地址，编译该模块，如果安装了图形化界面，自动mod_audio_fork模块。
https://gitee.com/rasonyang/drachtio-freeswitch-modules/blob/main/modules/mod_audio_fork/README.md
https://github.com/amigniter/mod_audio_stream

编译完成后，加载模块

freeswitch@MyFs> load mod_audio_fork 
+OK Reloading XML
+OK

mod_audio_fork对应的api为

uuid_audio_fork <uuid> start <wss-url> <mix-type> <sampling-rate> <metadata>

3.4 编写呼叫路由和呼叫脚本

呼叫路由

<extension name="11120-自定义ASR"><condition expression="^(11120)$" field="destination_number"><action application="python" data="handle-asr"/><action application="hangup"/></condition>
</extension>

呼叫脚本

#!/usr/local/python3/bin/python3
# -*- coding: utf-8 -*-
"""
处理语音识别结果
"""from freeswitch import *
import refsapi = API()def handler(session, args):# 接通session.answer()# 获取uuiduuid = session.getVariable("uuid")# 执行api,把该uuid的audio发送到asr地址的8000（上面的容器映射的端口）,单声道mono,采样率16kcmd = "uuid_audio_fork {} start ws://192.168.1.70:8000/ mono 16k".format(uuid)fsapi.executeString(cmd)# 超时时间session.sleep(30000)

3.5 esl捕获识别结果

连接esl，获取"Event-Subclass": "mod_audio_fork::json"事件，_body为识别内容。

{"Event-Subclass":	"mod_audio_fork::json","Event-Name":	"CUSTOM","Core-UUID":	"5788c155-5b0e-48c9-8ec8-accd46623a0b","FreeSWITCH-Hostname":	"localhost.localdomain","FreeSWITCH-Switchname":	"MyFs","FreeSWITCH-IPv4":	"192.168.1.20","FreeSWITCH-IPv6":	"::1","中间很多内容，略过......""variable_playback_seconds":	"1","variable_playback_ms":	"1000","variable_playback_samples":	"8000","Content-Length":	"144","Content-Length":	"144","_body":	"{\"text\":\"今天是205年8月2号然后我在测试一个语音识别程序我主要想看一下原始月的准确度\",\"timestamp\":1754120626173}"
}

3.6 其他处理

识别结果拿到后，解析body内容，可以在esl里编写程序，进行其他处理。

3.7 测试视频参考

下面为这个registry.cn-hangzhou.aliyuncs.com/pbx/asr-124m的测试结果

祝君成功，好运连连