Evolution of Voice Communication and the Design of the YTalk Call Platform Based on FreeSWITCH

1. Derivatives of Voice Communication

1. Traditional telephone implementation

In 1876 Alexander Graham Bell invented the first telephone, converting sound vibrations into electrical signals that traveled over a wire to a receiver, allowing two parties to hear each other; however, wired phones required a dedicated line to the operator, prompting the shift toward mobile telephony.

2. Birth of mobile phone

On April 3, 1973 Motorola released the Dyda TAC 8000X, the first true mobile phone, freeing voice communication from fixed lines. Mobile telephony evolved from 1G analog voice to 2G digital, and now to 5G, dramatically improving quality, capacity, and speed.

3. Development of Internet telephone

With rapid Internet growth, Voice over Internet Protocol (VOIP) emerged, enabling voice calls over any Internet connection without cellular towers. VOIP’s low development cost facilitated widespread adoption, exemplified by WeChat voice. In 2011, WebRTC opened browser‑based voice communication, allowing agents to call users directly from a PC.

2. YTalk Call Platform

1. Business background

YongQianGuan required both outbound and inbound calling capabilities. Existing third‑party solutions could not meet custom needs, so in 2019 development of the YTalk platform began, choosing FreeSWITCH for its extensibility and stability.

Manual outbound: dial a number from the UI and converse while handling business tasks.

IVR: play audio prompts for simple notifications such as repayment reminders.

Predictive outbound: calculate connection rates and seat availability in real time to balance dialing speed and agent idle time.

Intelligent outbound: use ASR and intent understanding to generate dynamic responses via TTS.

One‑click multi‑dial: dial a batch of numbers, connecting the first answered call to an agent.

Inbound: users call a 400 number and navigate menus or reach a live agent.

YTalk currently supports the following business lines:

Collection: manual, IVR, intelligent, multi‑dial, predictive, inbound.

Marketing: manual, IVR.

Insurance: manual, IVR, inbound.

Customer service: manual, inbound.

Alarm platform: IVR.

2. Introduction to FreeSWITCH

Traditional exchanges physically connected every pair of users, which is impractical at scale. Automatic exchanges (switches) route calls based on dialed numbers. In mobile networks, a device registers its number to nearby base stations, enabling the operator to locate the callee.

VOIP follows a similar registration model: a client registers its IP and port with a soft‑switch (FreeSWITCH), which then routes calls using this information.

FreeSWITCH is a pure‑software switch written in C, offering features such as seat registration, gateway integration, call initiation, audio streaming, recording, IVR menus, and DTMF handling.

3. SIP – Session Initiation Protocol

SIP is a text‑based application‑layer protocol used to create, modify, and terminate voice sessions. It can run over UDP, TCP, or WebSocket (as in WebRTC). SIP messages are visible in packet captures or browser dev tools.

Registration flow

The following SIP messages illustrate a typical registration sequence:

REGISTER sip:opensips-test.yangqianguan.com SIP/2.0
To: <sip:[email protected]>
From: <sip:[email protected]>;tag=jkmn1j53l4
Call-ID: nenr9bkuf95rqur5a9jjoj
Contact: <sip:[email protected]>;expires=600

SIP/2.0 401 Unauthorized
From: <sip:[email protected]>;tag=2c13hqji91
To: <sip:[email protected]>;tag=rKB1g622gN8Ze
Call-ID: nenr9bkuf95rqur5a9jjoj
WWW-Authenticate: Digest realm="opensips-test.yangqianguan.com", nonce="07577657-7c5d-451d-8c1b-6c1e1d1efd06", algorithm=MD5

REGISTER sip:opensips-test.yangqianguan.com SIP/2.0
To: <sip:[email protected]>
From: <sip:[email protected]>;tag=2c13hqji91
Call-ID: nenr9bkuf95rqur5a9jjoj
Authorization: Digest algorithm=MD5, username="8801", nonce="07577657-7c5d-451d-8c1b-6c1e1d1efd06", response="0570ba0a5efebbfb354634d549431123", cnonce="m83dogd7nimg"
Contact: <sip:[email protected];transport=wss>;expires=600

SIP/2.0 200 OK
From: <sip:[email protected]>;tag=2c13hqji91
To: <sip:[email protected]>;tag=Sv4Sj1K6Dyyja
Call-ID: nenr9bkuf95rqur5a9jjoj
Contact: <sip:[email protected]:5060;type=wss>;expires=220

Call flow

A typical outbound call from seat A (ID 8801) to mobile B (18812341234) proceeds through INVITE, 407 authentication challenge, 200 OK, media negotiation, and BYE termination. Key SIP messages are shown below.

INVITE sip:[email protected] SIP/2.0
To: <sip:[email protected]>
From: <sip:[email protected]>;tag=knutvg0g37
Call-ID: 370kqsd7pfmcdb8g2md5
Content-Type: application/sdp

o=- 476668281734765389 2 IN IP4 10.0.13.xxx
m=audio 58361 UDP/TLS/RTP/SAVPF 111 103 104 9 0 8 106 105 13 110 112 113 126

SIP/2.0 183 Session Progress
From: "8801" <sip:[email protected]:6977>
To: <sip:[email protected]:6977>
Call-ID: 8f9732c0-5346-123a-e59e-00163e142701
Content-Type: application/sdp

o=- 1624948324 1624948324 IN IP4 223.123.123.123
m=audio 10498 RTP/AVP 8 96

SIP/2.0 200 OK
From: "8801" <sip:[email protected]:6977>
To: <sip:[email protected]:6977>
Call-ID: 8f9732c0-5346-123a-e59e-00163e142701

BYE sip:[email protected]:5062;transport=udp SIP/2.0
To: <sip:[email protected]>;tag=6ggF0887eZmtS
From: <sip:[email protected]>;tag=q09i4pbimg
Call-ID: 370kqsd7pfmcdb8g2md5

4. Implementation

FreeSWITCH is written in C, while YTalk’s services are Java‑based. Interaction occurs via the Event Socket Library (ESL) over a persistent TCP connection on port 8021. Java sends commands (e.g., originate , bridge , park ) to control calls and subscribes to all events using event plain all . Important events include CHANNEL_CREATE, CHANNEL_PROGRESS, CHANNEL_ANSWER, CHANNEL_PARK, CHANNEL_HANGUP_COMPLETE, sofia::register, and sofia::unregister. Event payloads contain call IDs, IPs, timestamps, and other metadata, enabling state tracking and database updates.

3. Manual Dial Scenario and Application

Manual dialing is the most common outbound use case, especially in collection workflows where agents call borrowers to remind them of repayments. Agents open a case, click the masked phone number, and the system automatically initiates the call.

4. Current Usage

In production, peak outbound volume reaches 600,000 calls per day, with a baseline of 450,000. Most calls are automated (IVR, predictive), while manual dialing accounts for a smaller share. Future goals include scaling to one million daily calls, improving throughput while maintaining quality, and enhancing stability, high‑availability, monitoring, and configurability.