Voice

The ASAPP Voice Agent Desk includes web-based agent-assist services, which provide telephone agents with a machine learning and natural-language processing powered desktop. Voice Agent Desk augments the agent’s ability to respond to inbound telephone calls from end customers.

Voice Agent Desk augments the agents by allowing quick access to relevant customer information and provides actionable suggestions that ASAPP infers from the analysis of the ongoing conversation. The content, actions, and responses ASAPP provides to agents is meant to augment the agent’s ability to respond quickly and more effectively to end customers.

Voice Agent Desk interfaces with relevant customer applications to enable desired features.

The ASAPP Voice Agent Desk is not in the call-path but is more of an active listener, and uses two different integrations to provide the real-time augmentation:

• SIPREC - you enable SIP RECording on the customer Session Border Controllers (SBC) and route a copy of the media stream, call information, and metadata per session to ASAPP.
• CTI Events - ASAPP subscribes to telephony events of the voice agents via the CTI server (login, logout, on-hook, off-hook, etc.)

You associate and aggregate the media sessions and CTI events within the ASAPP solution and use them to power the agent augmentation features presented in Voice Agent Desk to the agents.

The ASAPP Voice Agent Desk solution provides agents with the real-time features that automate many of their repeatable tasks. Agents can use Voice Agent Desk for:

• The real-time transcript
• Conversation Summary - where agents add notes and structured data tags that ASAPP suggests as well as disposition the call during the interaction and once it is complete.
• Agents login to Voice Agent Desk via the customer’s SSO.
• Customer information (optional)
• Knowledge Base integration (optional)

Customer Current State Solution

ASAPP works with you to understand your current telephony infrastructure and ecosystem, including the type of voice work assignment platform/s and other capabilities available, such as SIPREC.

Solution Architecture

After the discovery of the customer’s current state is complete, ASAPP completes the architecture definition, including integration points into the existing infrastructure. You can deploy the ASAPP media gateways and media gateway proxies within your existing AWS instance or within ASAPP’s, providing additional flexibility and control.

Network Connectivity

ASAPP will determine the network connectivity between your infrastructure and the ASAPP AWS Virtual Private Cloud (VPC) based on the architecture, however, there will be secure connections deployed between your data centers and the ASAPP VPC.

Port Details

You can see ports and protocols in use for the Voice implementation depicted in the following diagram. These definitions provide visibility to your security teams for the provisioning of firewalls and ACL’s.

• SIP/SIPREC - TCP (5060, 5070-5072)
  • SBC to Media Gateway Proxies
  • SBC to Media Gateway/s
• Audio Streams - UDP <RTP/RTCP port range>
• CTI Event Feed - TCP <vendor specific>
• API Endpoints - TCP 443 In customer firewalls, you must disable the SIP Application Layer Gateway (ALG) and any ‘Threat Detection’ features, as they typically interfere with the SIP dialogs and the re-INVITE process.

Data Flow

The Voice Agent Desk Data Flow diagram illustrates the PCI Zone within the ASAPP solution. The customer SBC originates the SIPREC sessions and the media streams and sends them to ASAPP media gateways, which repackage the streams into secure WebSockets and sends them to the Voice Streamer within the PCI zone. ASAPP encrypts the data in transit and at rest.

The SBC does not typically encrypt the SIPREC sessions and associated media streams from the SBC to the ASAPP media gateways, but usually encapsulates them within a secure connection. You are responsible for the compliance/security of the network path between the SBC and the media gateways, in accordance with applicable customer policies.

SIPREC and CTI Correlation and Association

In order to be able to associate the correct audio stream and the correct agent and agent desktop, ASAPP must associate the audio session and the CTI events of the particular agent.

ASAPP assigns voice agents a unique Agent ID and adds it to the SSO profile as a custom attribute. ASAPP will then map this to the Agent ID within ASAPP.

You configure the SBCs to set a unique call identifier, such as UCID (Avaya) or GUID/GUCID (Cisco), etc. on inbound calls, which provides ASAPP the means to correlate the individual SIPREC stream with the CTI events of the correct agent.

The SBCs will initiate a SIPREC session INVITE for each new call. With SIPREC, the customer SBC and the ASAPP media gateway negotiate the media attributes via the SDP offer/answer exchange during the establishment of the session. The codec/s in use today are:

1. G.711
2. G.729

Traffic and load considerations:

• Total number of voice agents using ASAPP -<total agent count>
• Maximum concurrently logged in agents <max concurrent agent count>
• Maximum concurrent calls at each SBC pair -<max number of current offered calls to SBC/s>
• Maximum calls per second at each SBC pair -<max calls per second offered to the SBC>

Load Balancing for ASAPP Media Gateway Proxies

In order to distribute traffic across all of the media gateway proxies, the SBCs load balance the SIPREC dialogs to the ASAPP MG Proxies. To facilitate this, you configure the SBCs with a proxy list that provides business continuity and enables the fail-over to the next available proxy if one of the proxies becomes unavailable.

Session Recording Group Example:

The customer data center SBCs use different orders for the media gateway proxy list.

Data Center 1:

1. MG Proxy #1
2. MG Proxy #2
3. MG Proxy #3

Data Center 2:

1. MG Proxy #3
2. MG Proxy #2
3. MG Proxy #1

Media Failover and Survivability

Session Border Controller (SBC) to Media Gateways (MG) and Proxies

• Typically unencrypted signaling and audio through a secure connection/private tunnel
  • You can encrypt the traffic in theory, but the SBC has costs and scale limitations associated with encrypting traffic, as well as cost increases to MGs as you will need more instances.
• ASAPP accepts SIPREC dialogs, but initially sets SDP media to “inactive,” which pauses the audio while in the IVR and in queue.
  • The ASAPP media gateway will re-invite the session and re-negotiate the media parameters to resume the audio stream when the agent answers the call.

• SIP RFC handles some level of packet loss and re-transmissions but if the SIP signal is lost, the SIPREC dialog will be torn down and the media will no longer be sent.
• Media is sent via UDP.
  • No retransmissions so packet loss or disconnects result in permanent loss of the audio.
• Proxies are transactionally stateless.
• No audio is ever sent to/through proxies, all audio goes directly to media gateways.
  • Proxies are no longer in the signal path after the first transaction.
• If a proxy fails or is disconnected, SBCs can “hunt” or failover to the next proxy in it’s configuration.
  • No existing calls are impacted.
• If media gateways fail or are disconnected, the next SIP transaction will fail and the existing media stream (if resumed) will be sent via UDP to nothing (media is lost).
• Media gateways use regular SIP OPTIONS sent to static proxies that indicate if they are available and their current number of calls.
• Proxies use this active call load to evenly load balance to the least used media gateway.
  • As well as dynamically pick up when a media gateway is no longer available or new ones come online.
• Any inbound calls coming in over ISDN-PRI/TDM trunk facilities will not have associated SIPREC sessions, as these calls do not traverse the SBC.

Media Gateways to ASAPP Voice Streamers

• Secure websocket initiated per stream (2 per call) to the ASAPP Voice Streamer
• Media gateways do not store media, all processing is done in memory.
• Packet loss can be tolerated a little with TCP retransmissions.
• Buffer overrun audio data in the media gateway is purged instantly (per stream).
• If a secure websocket connection is lost, the media gateway will attempt a limited number of reconnections and then fail.
• If a voice streamer fails, a media gateway will reconnect to a new streamer.
• If a media gateway fails, the SIPREC stream is lost and the SBC can no longer send audio for that group of calls.

Integration

API Integration

Integration to existing customer systems enable ASAPP to call for information from those systems to present to the agent, such as:

• customer profile information
• billing history/statements
• customer product purchases
• Knowledge Base

Integration also enables ASAPP to push information to those systems, such as disposition notes and account changes/updates.

ASAPP will work with you to determine use cases for each integration that will add value to the agent and customer experience.

Custom Call Data from CTI Information

In many instances, CTI will carry end customer specific information about the end customer and the call. This may be in the form of User-to-User Information (UUI), Call Variables, Custom NamedVariables, or Custom KVList UserData. ASAPP uses this data to provide more information to agents and admins. It may contain information that provides customer identity information, route codes, queue information, customer authentication status, IVR interactions/ outputs, or simply unique identifiers for further data lookup from APIs. ASAPP extracts the custom fields and leverages the data in real-time to provide agents as much information as possible as part of the initial part of the interaction.

Each environment is uniquely different and ASAPP needs to understand what data is available from the CTI events to maximize relevant data to the agent and for voice intelligence processing.

Examples:

Avaya

UserToUserInfo: “10000002321489708161;verify=T;english;2012134581”

Cisco

CallVariable1:10000002321489708161
CallVariable7:en-us
user.AuthDetect:87

Genesys

userAccount:10000002321489708161
userLanguage:en
userFirstName:John

Twilio

<Parameter name=”FirstName” value=”John”/>
<Parameter name=”AccountNum” value=”10000002321489708161”/>
<Parameter name=”Language” value=”English”/>
<Parameter name=”VerficationStatus” value=”True”/>

SSO Integration

Single Sign-On (SSO) allows users to sign in to ASAPP using their existing corporate credentials. ASAPP supports Security Assertion Markup Language (SAML) 2.0 Identity Provider (IdP) based authentication. ASAPP requires SSO integration to support implementation.

To enable the SSO integration, the customer must populate and pass the Agent Login ID as a custom attribute in the SAML payload. Then, when a user logs in to ASAPP and authenticates via the existing SSO mechanism, the Agent Login ID value is then passed to ASAPP via SAML assertion for subsequent CTI event correlation.

The ASAPP Voice Agent Desk supports role-based access. You can define a specific role for each user that will determine their permissions within the ASAPP platform. For example, you can define the “app-asappagentprod” role in the Active Directory to send to ASAPP via SAML for those specific users that should have access to ASAPP Voice Agent Desk only. You can define multiple roles for an agent, such as access to Voice Agent Desk, Digital Agent Desk, and Admin Desk. You must define roles for voice agents and supervisors and include them in the SAML payload as a custom attribute.

The table below provides examples of SAML user attributes.

SAML Attribute Values	ASAPP Usage	Examples
Agent Login ID	Provides mapping of the customer telephony agent ID to ASAPP’s internal user ID.	user.extensionattribute1 or cti_agent_id
Givenname	Given name	user.givenname
Surname	Surname	user.surname
Mail	Email address	user.mail
Unique User Identifier	The User ID (authRepId); can be represented as an employee ID or email address.	user.employeeid or user.userprincipalname
PhysicalDeliveryOfficeName	Physical delivery office name	user.physicaldeliveryofficename
HireDate	Hire date attribute used by reporting.	HireDate
Title	Can be used for reporting.	Title
Role	The roles define what agents can see in the UI and have access to when they login.	user.role app-asappadminprod app-asappagentprod
Group	For Voice, this is only for reporting purposes. For digital chat this also can be used for queue management.	user.groups

Call Flows

Once an inbound Automatic Call Distribution (ACD) call is connected to an agent, the agent may need to transfer or conference the customer in with another agent/skill group. It is important to identify and document these types of call flows, when the transcript and customer data needs to be provided to another agent due to a change in call state. Then ASAPP will test these call scenarios as part of the QA and UAT testing process.

These scenarios include:

• Cold Transfers
  • The agent transfers the call to a queue (or similar) but does not stay on the call.
• Warm Transfers
  • The agent talks to the receiving agent prior to completing the transfer, in order to prepare the agent with the context of the call/customer issue.
• Conferences
  • The agent conferences in another agent or supervisor and remains on the call.
• Other
  • Customer call back applications or other unique call flows.

Speech Files for Model Training

To prepare for a production launch, ASAPP will train the speech models on the customer language and vocabulary, which will provide better transcription accuracy. ASAPP will use a set of customer call recordings from previous interactions.

You will need to provide ASAPP with a minimum of 1,000 hours of agent/customer dual-channel/speech separated media files in .wav format with a sample rate of 8000 and signed 16-bit Pulse-Code Modulation (PCM) in order for ASAPP to train the speech recognition models.

• ASAPP will set up an SFTP site in our PCI zone to receive voice media files from you. You will provide an SSH public key and ASAPP will configure the SFTP location within S3.
  • ASAPP prefers that you redact the PCI data from the provided voice recordings. Regardless, ASAPP will use its media redaction technology to remove sensitive customer data (Credit Card Numbers and Social Security Numbers) from the recordings to the extent possible. In addition to the default redaction noted above, ASAPP can customize redaction criteria per your requirements and feature considerations.
• The unredacted voice media files will remain within the PCI Zone.
• ASAPP will use a combination of automated and manual transcription to refine our speech models. Data that ASAPP shares with vendors goes through the redaction process described above and is transferred via secured mechanisms such as SFTP.

Non-Production Lower Environments

As part of our implementation strategy, ASAPP will implement two lower environments for testing (UAT and QA) by both ASAPP and customer resources. It is important that the lower environments do not use production data, including the audio data, as it may contain PCI information or other customer information that you should not expose to the lower environments.

You can implement lower environments using a lab environment, or a production environment.

When using the production infrastructure to support the lower environments, ASAPP separates production traffic from the lower environment traffic. The lower environments will have dedicated inbound numbers and routing that will allow them to be isolated and provide the ability for ASAPP and the customer teams to fully test using non-production traffic.

As part of the environment’s buildout, ASAPP will need a way to initiate and terminate test calls. The ASAPP team will use the same soft-client and tools used by agents to login as a voice agent, answer inbound test calls, and simulate the various call flows used within the customer contact center.

ASAPP proposes customers allocate two Direct Inward Dialing (DID)/ Toll Free Number (TFN) numbers, one for each of the two different test environments.

• Demo Environment - A lower environment used by both ASAPP and customers.
• Preprod Environment - A lower environment used by ASAPP QA for testing.

At the SBC level, you should configure the Demo and Preprod DID numbers with their own Session Recording Server (SRS), unique from the production SRS configuration. This will allow the test environments to always have SIPREC turned on, but not send excess/production traffic to ASAPP. This also allows the test environments to operate independently of production. With Oracle/Acme, you can accomplish this with session agents. For Avaya SBCE, you can accomplish this with End Point Flows.

ASAPP will have a separate set of media gateways and media gateway proxies for each environment to ensure traffic and data separation.

The lower environments (not PCI compliant) are for testing only and will not receive actual customer audio. The production environment is where ASAPP transcribes and redacts the audio in a PCI zone.

Appendix A - Avaya Configuration Details

This section provides specific configuration details for the solution that leverages Avaya telephony infrastructure.

Avaya Communication Manager

• Set Avaya Internet Protocol - Private Branch Exchange (IP-PBX) SIP trunks to ‘shared’ to ensure the UCID is not reset by the PBX.
  • Change trunk-group x -> page 3 -> UUI Treatment:shared
• Set SendtoASAI parameter to ‘yes.’
  • Change system-parameters features -> page 13 -> Send UCID to ASAI? Y
  • Add ASAPP voice agents to a new skill, one that is not used for queuing or routing.
  • Configure AES to monitor the new skill.
    • ASAPP will use the cstaMonitorDevice service to monitor the ACD skill.
    • ASAPP may also call cstaMonitorCallsViaDevice if more call data is needed.

Avaya AES TSAPI configuration

• Networking -> Ports -> TSAPI Ports
  • Enabled
  • TSAPI Service Port (450)
    • Firewalls will also need to allow these ports.

Connection Type	TCP Min Port	TCP Max Port
unencrypted/TCP	1050	1065
encrypted/TLS	1066	1081

• AES link to ASAPP connection provisioning
  • Provisioning of new ASAPP Voice skill for monitoring.

Appendix B - Cisco Configuration Details

This section provides specific configuration details for the solution that leverages Cisco telephony infrastructure.

Cisco CTI Server configuration

• ASAPP will connect with the CTI_SERVICE_ALL_EVENTS
  • You will need the Preferred ClientID (identifier for ASAPP) and ClientPassword (if not null) to send the OPEN_REQ message.
• Ports 42027 (side A) and 43027 (side B)
  • Instance number if not 0 will increase these ports
  • Firewalls will also need to allow these ports
• CallVariable1-10 Definitions/usages
• Custom NamedVariables and NamedArrays Definitions/usages
• Events currently used by ASAPP:
  • OPEN_REQ
  • OPEN_CONF
  • SYSTEM
  • AGENT_STATE
  • AGENT_PRE_CALL
  • BEGIN_CALL
  • CALL_DATA_UPDATE
  • CALL_CLEARED
  • END_CALL

Appendix C - Oracle (Acme) Session Border Controller

In order to provide the correlation between the SIPREC session and specific CTI events, ASAPP will use the following approach:

• Session Border Controller
  • Configure the SBC to create an Avaya UCID (universal call identifier) in the SIP header.
    • UCID generation is a native feature for Oracle/Acme Packet session border controller platforms.
      • Oracle SBC UCID Admin
• In the Oracle (Acme Packet) SBCs, load balancing across the ASAPP Media Gateway Proxies requires the use of static IP addresses versus the use of dynamic hostnames.
  • SBC Settings for Media Gateway Proxies - Production and Lower Environments:
    • Transport = TCP
    • SIP OPTIONS = disabled
    • Load Balancing strategy = “hunt”
    • Session-recording-required = disabled
    • Port = 5070

Glossary

Term	Acronym	Definition
Automated Speech Recognition	ASR	The service that converts speech (audio) to text.
Automatic Call Distributor	ACD	A telephony system that automatically receives incoming calls and distributes them to an available agent. Its purpose is to help inbound contact centers sort and manage large volumes of calls to avoid overwhelming the team.
Computer Telephony Integration	CTI	The means of linking a call center’s telephone systems to a business application. In this case, ASAPP is monitoring agents and receives call state event data via CTI.
Direct Inward Dialing	DID	A service that allows a company to provide individual phone numbers for each employee without a separate physical line.
Globally Unique IDentifier	GUID	A numeric label used for information in communications systems. When generated according to the standard methods, GUIDs are, for practical purposes, unique. Also known as Universally Unique IDentifier (UUID)
Internet Protocol Private Branch Exchange	IP-PBX	A system that connects phone extensions to the Public Switched Telephone Network (PSTN) and provides internal business communication.
Media Gateway	MG	Entry point for all calls from Customer. Receives and forwards SIP and audio data.
Media Gateway Proxy	MGP	SIP Proxy, used for SIP signaling to/from customer SBC.
Payment Card Industry Data Security Standard	PCI DSS	Payment card industry compliance refers to the technical and operational standards that businesses follow to secure and protect credit card data provided by cardholders and transmitted through card processing transactions.
Payment Card Industry Zone	PCI Zone	PCI Level I Certified environment for cardholder data and other sensitive customer data storage (Transport layer security for encryption in transit, encryption at rest, access tightly restricted and monitored).
Pulse-Code Modulation	PCM	Pulse-code modulation is a method used to digitally represent sampled analog signals. It is the standard form of digital audio in digital telephony.
Security Assertion Markup Language	SAML	An open standard for exchanging authentication and authorization data between an identity provider and a service provider.
Session Border Controller	SBC	SIP-based voice security platform; source of the SIPREC sessions to ASAPP.
Session Description Protocol	SDP	Used between endpoints for negotiation of network metrics, media types, and other associated properties, such as codec and sample size.
Session Initiation Protocol Application-Level Gateway	SIP ALG	A firewall function that enables the firewall to inspect the SIP dialog/s. This function should be disabled to prevent SIP dialog interruption.
Session Initiation Protocol Recording	SIPREC	IETF standard used for establishing recording sessions and reporting of the metadata of the communication sessions.
Single Sign On	SSO	Single sign-on is an authentication scheme that allows a user to log in with a single ID and password to any of several related, yet independent, software systems.
Toll-Free Number	TFN	A service that allows callers to reach businesses without being charged for the call. The called person is charged for the toll-free number.
Telephony Services API	TSAPI	Telephony server application programming interface (TSAPI) is a computer telephony integration standard that enables telephony and computer telephony integration (CTI) application programming.
Universal Call IDentifier	UCID	UCID assigns a unique number to a call when it enters that call center network. The single UCID can be passed among platforms, and can be used to compile call-related information across platforms and sites.
User to User Information	UUI	The SIP UUI header allows the IVR to insert information about the call/caller and pass it to downstream elements, in this case, Communication Manager. The UUI information is then available via CTI.
Voice Streamer	VS	Receives SIP and audio data from MG. Gets the audio transcribed into text through the ASR and sends that downstream.

Assistant

Responses are generated using AI and may contain mistakes.