TECHNICAL ARCHITECTURE

How Does Phone AI Work?

Learn how AI answers incoming calls and what environment preparation is needed before implementation.

Overall Flow

From call connection to recording, incoming calls go through the following three steps.

1
Connect
Route PSTN calls through IP-PBX or VoIP to connect with TS Cloud Voice AI
2
Respond
Gemini Live directly processes voice for natural, real-time conversational responses
3
Record
Call content and recordings are automatically saved to the client's dedicated Google Cloud project for analysis
STEP 1

Connecting Calls to Cloud-Based Voice AI

PSTN phone standards and internet protocols are two completely different communication standards, so format conversion is required before connecting to Google AI in the cloud. Depending on your office's existing environment, there are two main scenarios.

ENVIRONMENT A

Already Using IP-PBX or Cloud Phone

An IP-PBX is essentially a switchboard that converts phone lines to internet protocol. If all phones in your office are connected via ethernet cables, you almost certainly have an IP-PBX or cloud phone system.

In this case, simply configure call forwarding for the desired numbers to TS Cloud Voice AI within your system to complete the connection.

ENVIRONMENT B

Using Standard Phone Lines

In this case, the phone line needs to be converted to internet protocol or undergo format conversion. There are three main options available.

1

Apply for an Internet Phone Number (VoIP Number)

Apply for a new VoIP number from your telecom provider and configure forwarding to TS Cloud Voice AI. For inbound call purposes, a VoIP number works perfectly well.

2

Convert Your Existing Number to Internet Protocol

Convert your current phone number to internet format. Contact your current telecom provider to arrange this.

3

Install a VoIP Gateway on Your Existing Phone Line

Connect a VoIP Gateway device to your existing phone line for SIP conversion. A VoIP Gateway is a device that converts phone signals to internet protocol, available for approximately NT$3,000 per line.

Options 1 and 2 require cooperation from local telecom providers. Due to anti-fraud and identity verification regulations, telecom providers are generally conservative about allowing external system SIP integration, which may not be permitted in practice. Before applying, please confirm with your telecom provider whether they support call forwarding to third-party systems (SIP).

If your provider cannot accommodate this, Option 3 (VoIP Gateway) does not require telecom provider cooperation — simply install the device yourself to complete the conversion. This is the most straightforward approach. When purchasing equipment, please ensure the VoIP Gateway meets the following required specifications.

Click to expand: VoIP Gateway Technical Specifications
  • SIP URI Dialing
  • Digest Authentication / Username-Password Authentication
  • NAT Traversal Configuration
  • SIP Keepalive
  • STUN or External IP Configuration
  • RTP NAT Traversal
STEP 2

AI Understands and Responds to Calls

Whether it's Gemini or ChatGPT, most AI models are fundamentally "text-only brains." Traditional voice customer service required an extra "translation layer" around the AI for bidirectional voice-to-text conversion, which inevitably caused delays.
To break this limitation, Google launched the Gemini Live model with "native voice processing" capabilities — it can directly understand speech and respond with a human voice, delivering zero-latency real interactions!

Traditional Voice Processing

User Voice STT (Speech-to-Text) LLM Processing TTS (Text-to-Speech) Voice Response

⚠️ Pain Points of Traditional Approach

  • Translation Lag: Multiple conversion steps cause noticeable response delays.
  • Robotic Sound: Rigid text-to-speech conversion lacks the natural warmth of human speech.
💡

Smart AI on the Front End, Resilient Infrastructure on the Back End

Beyond Gemini Live's smooth conversation capabilities, TS Cloud's infrastructure is built on Google Cloud architecture (GKE). This ensures that even when multiple calls come in simultaneously, the AI processing core remains stable and uninterrupted.

STEP 3

Recording Content After Call Ends

After each conversation, records and audio files are stored in the client's dedicated Google Cloud project. Data is stored in Google's data center, meeting military-grade security standards to provide businesses with the most secure, compliant data protection environment.

BigQuery (Conversation Data & Analytics)

Real-Time Call Transcripts

During voice interactions, the system converts conversations to text and automatically stores them in a structured database. (Note: Speech recognition may occasionally be affected by accents or background noise — original recordings can be used to verify actual conversations.)

AI Performance Dashboard

Integrates with Data Studio to transform monthly call volume, average duration, and other metrics into visual charts, giving you real-time visibility into AI service capacity.

Data Analytics & Utilization

All call content becomes your company's digital asset, enabling deeper analysis and diverse applications to meet various business development needs.

Cloud Storage (Recording Preservation & Utilization)

The Most Authentic Verification Evidence

Original audio files are encrypted and stored. When issues or customer disputes arise, you can access the "original recording" of AI conversations anytime as your most reliable evidence.

Advanced Analysis & Applications

Properly preserved voice data supports post-processing through AI tools like Gemini, flexibly applied to interview scoring, business conversation analysis, and other scenarios to maximize data value.