BSS Digital2Speech

The main articles are:

• Omni channel
• Voice biometrics

Digital2Speech is a platform for automating omnichannel service of customers who have contacted the support service.

2024

Implementation of generative intelligence

BSS has expanded its voice analytics capabilities by analyzing dialogues with generative intelligence. The company's customers and partners have the tools to take the quality of service to the next level and promote business growth. The developer announced this on June 25, 2024.

The BSS team continues to innovate in generative artificial intelligence into their products. Earlier, the InKnowledge knowledge base from L2U, built into the Digital2Speech omnichannel dialog platform, received an intelligent search implemented using the GPT neural network and RAG technology.

The application of large language models (LLMs) in speech analytics has enriched it with new advanced functionality:

• Summarization - a brief summary of the dialogues
• Autoclassification of cases
• Auto-Service Quality Assessment
• Audit and identification of achievements and areas of employee growth
• Determination of emotional temperature and mood of operator and client
• Making recommendations for improvements
• Interactive analysis in natural dialogue mode with GPT service

GPT analysis allows you to build a continuous cycle of improvements in service and sales processes, reducing the cost of service, increasing its quality and increasing sales of products and services.

We have all the necessary tools and expertise for the implementation and deployment of GPT analysis in communication with clients, commented Alexander Krushinsky, Director of Voice Digital Technologies at BSS.

Migrating to Zipformer Architecture

BSS has improved the performance of its Automatic Speech Recognition (ASR) module used in the Digital2Speech dialog platform. The updated version of ASR is more than three times faster, while maintaining high quality recognition. The company announced this on June 7, 2024.

Thanks to a recent update, the response rate of the voice robot has more than tripled. This means that the voice assistant now responds to user requests noticeably faster, so interaction with it has become smoother and more natural. Despite the significant increase in speed, the quality of speech recognition not only did not suffer, but also improved.

The main factor that made it possible to implement these improvements and achieve the result was the transition to the neuronets Zipformer architecture. This architecture significantly speeds up the recognition process and increases its accuracy due to the reduced number of layers of the neural network, special compression methods data and a number of other mechanics. The company previously used Conformer technology.

The updated architecture can significantly reduce hardware requirements and save on computing power by increasing efficiency.

The optimized solution is tailored for use in the customer's loop. Therefore, the updated ASR can still be run both on GPU cards for maximum performance and on more affordable x86 servers.

The response speed of the voice robot is one of the three most important characteristics by which users evaluate the "humanity" and "pleasantness" of the voice assistant. And we were able to at least three times speed up the recognition of the subscriber's question, making interaction with our voice robot even more comfortable. Zipformer is an effective architecture for speech recognition tasks that appeared only a year ago. I am glad that we were one of the first to be able to present it to our customers and prepare a reliable Enterprise-level solution, "commented Alexander Krushinsky, Director of Voice Digital Technologies at BSS.

Improve speech recognition by improving diarization

BSS has improved the quality of speech recognition through improved diarization. The developer announced this on February 29, 2024.

Improving the technology of determining speakers in monoscopes, allows you to obtain a better transcription of conversations for subsequent analysis.

During the analysis of audio recordings, there may be problems with the confusion of the source of speech - who says what. This affects the results of speech analytics, and in general, the efficiency of virtual assistants.

Diarization allows you to separate audio streams when communicating and determine who exactly speaks at a particular moment in the dialogue. It is used to analyze monoscopes and recordings from audio books, as well as to improve the quality of recording recognition from directed stereo microphones when extraneous sounds penetrate.

For February 2024, the BSS developers have concentrated on three main areas for improving diarization, which will eliminate speech recognition flaws and increase the efficiency of speech analytics.

First, a channel classifier was added. Dialogs with customers are usually started by operators. Therefore, the first channel is usually assigned to the operator, which previously led to confusion, since the operator did not always start the conversation. The customer and operator speech records are now defined by two models. The first separates the replicas of speakers by biometric signs (voice). The second is understood by the characteristic vocabulary by which of the speaking operator and who is the client. As a result, the channel ownership is determined correctly.

Secondly, audio recordings pass through channel verification. For example, the courier delivers orders during the day, and all this time there is an audio recording. For analysis, only those fragments where the courier talks to customers are important. It is these parts of the recording using channel verification technology that are isolated from the entire audio stream for dirization. This can significantly reduce the burden on the IT infrastructure and reduce the processing and analysis time of audio.

Thirdly, the quality of the technology to eliminate duplication of replicas has been significantly improved. Usually consultations in banks are recorded using two directional microphones: one records a client, the other - a consultant. But often the client's speech is recorded by the consultant's microphone and vice versa. This leads to duplication of information in the text decryption and complicates the analysis of the dialogue using speech analytics. It was possible to achieve the elimination of duplicate replicas in the audio file by connecting both audio tracks into one stereo recording. Only then is the audio file transcribed. This tool has already shown its effectiveness in several banks.

{{quote 'Diarization improves the quality of texts during transcription and significantly improves the quality of speech when used with the recognition system. This technology is used by many developers, but we were able to significantly improve it and continue to constantly develop it. It is in demand, as it makes speech technologies more accessible. Many companies record conversations with customers on outdated equipment in mono format. Dialization allows such customers to implement the same speech analytics without the need to replace expensive equipment, "commented Alexander Krushinsky, Director of the Voice Digital Technologies Department at BSS. }}

As part of the contact center management solution

and SATEL BSS created a software product that allows you to optimize the work with customer requests in all digital communication channels. This system helps to speed up the maintenance and operation of operators by contact center reducing and automating routine operations. The solution is based on the kernel omnicanal contact center RTU-Atmosphere and intelligent dialog platform Digital2Speech. SATEL announced this on January 17, 2024. Подробнее #.2A2024:_.D0.92_.D1.81.D0.BE.D1.81.D1.82.D0.B0.D0.B2.D0.B5_.D1.80.D0.B5.D1.88.D0.B5.D0.BD.D0.B8.D1.8F_.D0.B4.D0.BB.D1.8F_.D1.83.D0.BF.D1.80.D0.B0.D0.B2.D0.BB.D0.B5.D0.BD.D0.B8.D1.8F_.D0.BA.D0.BE.D0.BD.D1.82.D0.B0.D0.BA.D1.82-.D1.86.D0.B5.D0.BD.D1.82.D1.80.D0.BE.D0.BC здесь.

2022

Inclusion in the "Financial Sector Software Interchangeability Catalog"

The company BSS on August 11, 2022 announced the inclusion of its own solutions for, and in the "Interchangeability Catalog remote banking speech technologies information security for the ON financial Sector." In particular, from BSS omnicanal , the Digital2Speech dialog platform is included in the Catalog. More. here

Functionality to optimize the primary markup process

On June 1, 2022, BSS announced the completion of the first stage of the project for the development of dialogue technologies based on artificial intelligence and machine learning with grant support from the Russian Information Technology Development Fund (RFRIT).

BSS Digital2Speech

According to the company, the goal of the project is to create Russian a competitive vendor solution based on speech technologies, which will become the basis. import-independent Its task is to provide business and state structures Russia CIS other countries with a complex of speech services using artificial intelligence and machine learning.

The full stack of native technologies will be enhanced by the use of an end-to-end (E2E) approach in speech recognition and few-shot learning in natural language understanding. This will reduce the time and resources to create application solutions in the field of dialog technologies (text and voice intelligent assistants, text and voice analytics, etc.).

At the first stage of the project, opportunities were implemented within the framework of the omnichannel Digital2Speech dialogue platform regarding speech recognition, speech analytics and dialogue maintenance tools. In particular:

• Voice-to-text recognition models based on artificial intelligence technologies, which allows optimizing the quality of speech recognition with the predominance of vocabulary of requests to government agencies through the use of an adapted e2e ASR model (gives more proper accuracy of speech recognition and understanding of natural language) in voice virtual assistant projects.
• Functionality that allows you to optimize the process of primary markup and verification of existing replica markup before training classifier models using set splitting into semantic clusters.
• Capabilities that allow voice analytics users to quickly filter and evaluate analyzed dialogs with setting filtering and evaluation values, specifying conditions for quantitative-time and lexical-semantic parameters, as well as the ability to build and upload reports based on the results of speech analysis.
• Free dialogue mechanics, the use of which allows you to optimize the architecture of developing service scripts, increase the flexibility of applications for processing client requests by virtual assistants: processing at the level of functional elements of slots, exceptions, changing the main intention without the need to write code.

During the already started second stage, the implementation of the updated functionality of BSS speech solutions continues: improving speech recognition, dialogue support tools and speech analytics. The end of the second stage is planned in January 2023.

Integration with InKnowledge L2U

On May 16, 2022, BSS announced that the InKnowledge Knowledge Base from L2U, a BSS partner, is built into the technology stack of the Digital2Speech omnichannel dialog platform. It expands the capabilities of the platform, acting as a single source of information that ensures the synchronization and maintenance of the relevance of information among integrated "consumers" of data (CRM, sites, portals, AWS of contact center operators, etc.).

This domestic solution, according to the company, is not inferior in quality, performance and functionality to current foreign storage and knowledge management systems such as Confluence, Notice, SharePoint, KMS Lighthouse and others.

The main difference L2U InKnowledge is that this Knowledge Base (KB) provide omnichannel knowledge that supports work with operators, robots, as well as in a self-service format. It can serve as a platform for organizing a single company reference center and information support for customer service in all communication channels.

Features of the InKnowledge L2U:

• Included in the register of domestic software (certificate No. 2021617316).
• It is quickly installed, including in the closed loop of the customer.
• Flexible it is integrated infrastructure with the customer.
• Scales easily, improving customer support, Service Desk, marketing, HR, and more.
• No-code. Allows you to independently configure interfaces from a set of widgets for different user tasks without programming.
• Allows you to quickly migrate content from different systems.

InKnowledge L2U capabilities:

• Store structured and unstructured information.
• Easily enter information into data structures.
• BZ Article Lifecycle Management.
• Supports multiple information domains in one installation.
• Flexible search of information in BZ with different variations of search results output.
• Flexible user interface.

"Users of our omnichannel dialog platform Digital2Speech appreciate the capabilities of the InKnowledge knowledge base. They note that in the current geopolitical situation, this Russian development is the optimal alternative to foreign counterparts. Knowledge base capabilities allow you to quickly integrate the product into the company's infrastructure, achieve lower operating costs, easily scale the solution, focusing on business development and growth, "- noted the company's sales director Vitaly Pateshman.

"By acquiring a packaged solution that includes a virtual assistant that is integrated with a full knowledge base, the customer reduces implementation costs and receives a comprehensive offer with an omnichannel knowledge source that closes both the text and voice channels of interaction with customers and provides information from a single source to robots, operators and any other consumers of knowledge," - emphasized L2U CEO Kirill Kuznetsov.

2021: Modules Digital2Speech

Omnicanal dialog platform is Digital2Speech based on proprietary technologies. BSS

As of February 2021, the Digital2Speech consists of a number of integrated modules operating on a single core:

• A module for identifying customer intentions, including extracting key attributes from speech, including a training tool and setting up models.
• Self-learning virtual assistant.
• Speech recognition, including a training tool and model settings.
• A voice biometrics module that includes identification and verification of the speaker throughout the conversation.
• Speech analytics module, including recognition of the speaker's moods and satisfaction with the product management and configuration tool.
• A tool for creating and managing dialog scripts and self-service services.
• Predictive synthesis of text into live speech.

The platform allows you to serve customers through all channels, including the most natural for a person - voice. This uses all the possibilities of omnichannel - you can start a dialogue in the voice channel and continue in a text chat and vice versa. Easy-to-use tools allow you to quickly change and create new dialogue scenarios without involving specialists.

2020

Passing testing for compatibility with the Rock-R

On August 13, 2020, SCALA-R announced that three systems related to the organization of contact centers were successfully tested for compatibility with the SCALA-R hyperconverged platform: a contact center manufactured by STC PROTEI LLC, a dialogue platform Digital2Speech manufactured by BSS, a workload management system for the ARGUS WFM CC contact center manufactured by STC ARGUS LLC. Read more here.

Description of Digital2Speech capabilities

According to information for June 2020, Digital2Speech is a single omnichannel platform based on artificial intelligence, which includes speech, meaning and emotion recognition services, as well as speech analytics and voice biometrics. Voice assistants and chatbots of the platform respond to customer requests, which reduces the load on the contact center and provides support for calls around the clock. If robots cannot answer the question, they transfer calls to company employees.

Neural networks allow robots to constantly improve: they take on more and more tasks, analyze intonations, timbre and pauses, which increases the accuracy of understanding requests. In the case of typical requests, the Digital2Speech is able to provide complete automation of interaction with customers. This is especially true during peak periods - customer service occurs without a long wait on the line, and only a part of calls are transferred to operators.

Digital2Speech allows you to scale the client service to the organization of full omnichannel service with the capabilities of voice control of applications on mobile devices and websites. The IT solution helps to transfer communications from the voice channel to the text channel and vice versa and has authentication and voice identification functions. A suite of flexible tools allows companies to independently train robots and adapt to different industries.

AI robots take into account the features, technical terminology, characteristic questions and typical dialogues of a particular organization. The platform can quickly adapt to the narrow requirements of the company and its customers, including on the basis of the dialogue database provided by the company, which makes it applicable to any industry of commercial and state-owned enterprises, V2V- and B2C segments.

Digital2Speech from BSS is safe: the solution was developed in accordance with the principles and rules for processing personal data provided for by federal law (152-FZ).