Your contact middle connects your small business to your group, enabling clients to order merchandise, callers to request help, shoppers to make appointments, and way more. Every dialog with a caller is a chance to study extra about that caller’s wants, and the way effectively these wants had been addressed through the name. You possibly can uncover insights from these conversations that enable you handle script compliance and discover new alternatives to fulfill your clients, maybe by increasing your providers to handle reported gaps, bettering the standard of reported drawback areas, or by elevating the shopper expertise delivered by your contact middle brokers.
Contact Lens for Amazon Join offers name transcriptions with wealthy analytics capabilities that may present these sorts of insights, however you could not at the moment be utilizing Amazon Join. You want an answer that works together with your present contact middle name recordings.
Amazon Machine Studying (ML) providers like Amazon Transcribe Name Analytics and Amazon Comprehend present feature-rich APIs that you should use to transcribe and extract insights out of your contact middle audio recordings at scale. Though you could possibly construct your personal customized name analytics resolution utilizing these providers, that requires time and assets. On this publish, we introduce our new pattern resolution for publish name analytics.
Answer overview
Our new pattern resolution, Publish Name Analytics (PCA), does a lot of the heavy lifting related to offering an end-to-end resolution that may course of name recordings out of your present contact middle. PCA offers actionable insights to identify rising traits, determine agent teaching alternatives, and assess the overall sentiment of calls.
You present your name recordings, and PCA mechanically processes them utilizing Transcribe Name Analytics and different AWS providers to extract beneficial intelligence comparable to buyer and agent sentiment, name drivers, entities mentioned, and dialog traits comparable to non-talk time, interruptions, loudness, and speak pace. Transcribe Name Analytics detects points utilizing built-in ML fashions which were skilled utilizing hundreds of hours of conversations. With the automated name categorization functionality, it’s also possible to tag conversations primarily based on key phrases or phrases, sentiment, and non-talk time. And you may optionally redact delicate buyer knowledge comparable to names, addresses, bank card numbers, and social safety numbers from each transcript and audio information.
PCA’s net consumer interface has a house web page displaying all of your calls, as proven within the following screenshot.
You possibly can select a report to see the small print of the decision, comparable to speech traits.
You too can scroll right down to see annotated turn-by-turn name particulars.
You possibly can seek for calls primarily based on dates, entities, or sentiment traits.
You too can search your name transcriptions.
Lastly, you may question detailed name analytics knowledge out of your most popular enterprise intelligence (BI) instrument.
PCA at the moment helps the next options:
- Transcription
- Analytics
- Caller and agent sentiment particulars and traits
- Speak and non-talk time for each caller and agent
- Configurable Transcribe Name Analytics classes primarily based on the presence or absence of key phrases or phrases, sentiment, and non-talk time
- Detects callers’ essential points utilizing built-in ML fashions in Transcribe Name Analytics
- Discovers entities referenced within the name utilizing Amazon Comprehend commonplace or customized entity detection fashions, or easy configurable string matching
- Detects when caller and agent interrupt one another
- Speaker loudness
- Search
- Search on name attributes comparable to time vary, sentiment, or entities
- Search transcriptions
- Different
- Detects metadata from audio file names, comparable to name GUID, agent’s title, and name date time
- Scales mechanically to deal with variable name volumes
- Bulk masses massive archives of older recordings whereas sustaining capability to course of new recordings as they arrive
- Pattern recordings so you may shortly check out PCA for your self
- It’s straightforward to put in with a single AWS CloudFormation template
That is just the start! We count on so as to add many extra thrilling options over time, primarily based in your suggestions.
Deploy the CloudFormation stack
Begin your PCA expertise through the use of AWS CloudFormation to deploy the answer with pattern recordings loaded.
- Use the next Launch Stack button to deploy the PCA resolution within the
us-east-1
(N. Virginia) AWS Area.
The supply code is accessible in our GitHub repository. Comply with the instructions within the README to deploy PCA to further Areas supported by Amazon Transcribe.
- For Stack title, use the default worth,
PostCallAnalytics
. - For AdminUsername, use the default worth, admin.
- For AdminEmail, use a legitimate e-mail tackle—your short-term password is emailed to this tackle through the deployment.
- For loadSampleAudioFiles, change the worth to
true
. - For EnableTranscriptKendraSearch, change the worth to
Sure, create new Kendra Index (Developer Version)
.
When you’ve got beforehand used your Amazon Kendra Free Tier allowance, you incur an hourly value for this index (extra data on value later on this publish). Amazon Kendra transcript search is an non-compulsory function, so for those who don’t want it and are involved about value, use the default worth of No.
- For all different parameters, use the default values.
If you wish to customise the settings later, for instance to use customized vocabulary to enhance accuracy, or to customise entity detection, you may replace the stack to set these parameters.
The primary CloudFormation stack makes use of nested stacks to create the next assets in your AWS account:
The stacks take about 20 minutes to deploy. The primary stack standing exhibits as CREATE_COMPLETE when every part is deployed.
Set your password
After you deploy the stack, that you must open the PCA net consumer interface and set your password.
- On the AWS CloudFormation console, select the primary stack,
PostCallAnalytics
, and select the Outputs tab. - Open your net browser to the URL proven as
WebAppURL
within the outputs.
You’re redirected to a login web page.
- Open the e-mail your obtained, on the e-mail tackle you supplied, with the topic “Welcome to the Amazon Transcribe Publish Name Analytics (PCA) Answer!”
This e-mail incorporates a generated short-term password that you should use to log in (as consumer admin) and create your personal password.
- Set a brand new password.
Your new password will need to have a size of not less than eight characters, and include uppercase and lowercase characters, plus numbers and particular characters.
You’re now logged in to PCA. Since you set loadSampleAudioFiles
to true, your PCA deployment now has three pattern calls pre-loaded so that you can discover.
Non-compulsory: Open the transcription search net UI and set your everlasting password
Comply with these further steps to log in to the companion transcript search net app, which is deployed solely while you set EnableTranscriptKendraSearch
while you launch the stack.
- On the AWS CloudFormation console, select the primary stack,
PostCallAnalytics
, and select the Outputs tab. - Open your net browser to the URL proven as
TranscriptionMediaSearchFinderURL
within the outputs.
You’re redirected to the login web page.
- Open the e-mail your obtained, on the e-mail tackle you supplied, with the topic “Welcome to Finder Net App.”
This e-mail incorporates a generated short-term password that you should use to log in (as consumer admin).
- Create your personal password, similar to you already did for the PCA net software.
As earlier than, your new password will need to have a size of not less than eight characters, and include uppercase and lowercase characters, plus numbers and particular characters.
You’re now logged in to the transcript search Finder software. The pattern audio information are listed already, and prepared for search.
Discover publish name analytics options
Now, with PCA efficiently put in, you’re able to discover the decision evaluation options.
Dwelling web page
To discover the house web page, open the PCA net UI utilizing the URL proven as WebAppURL
in the primary stack outputs (bookmark this URL, you’ll use it typically!)
You have already got three calls listed on the house web page, sorted in descending time order (most up-to-date first). These are the pattern audio information.
The calls have the next key particulars:
- Job Title – Is assigned from the recording audio file title, and serves as a singular job title for this name
- Timestamp – Is parsed from the audio file title if potential, in any other case it’s assigned the time when the recording is processed by PCA
- Buyer Sentiment and Buyer Sentiment Development – Present the general caller sentiment and, importantly, whether or not the caller was extra constructive on the finish of the decision than at first
- Language Code – Reveals the required language or the mechanically detected dominant language of the decision
Name particulars
Select probably the most just lately obtained name to open and discover the decision element web page. You possibly can overview the decision data and analytics comparable to sentiment, speak time, interruptions, and loudness.
Scroll right down to see the next particulars:
- Entities grouped by entity sort. Entities are detected by Amazon Comprehend and the pattern entity recognizer string map.
- Classes detected by Transcribe Name Analytics. By default, there are not any classes; see Name categorization for extra data.
- Points detected by the Transcribe Name Analytics built-in ML mannequin. Points succinctly seize the primary causes for the decision. For extra data, see Concern detection.
Scroll additional to see the turn-by-turn transcription for the decision, with annotations for speaker, time marker, sentiment, interruptions, points, and entities.
Use the embedded media participant to play the decision audio from any level within the dialog. Set the place by selecting the time marker annotation on the transcript or through the use of the participant time management. The audio participant stays seen as you scroll down the web page.
PII is redacted from each transcript and audio—redaction is enabled utilizing the CloudFormation stack parameters.
Search primarily based on name attributes
To strive PCA’s built-in search, select Search on the prime of the display. Below Sentiment, select Common, Buyer, and Unfavorable to pick out the calls that had common adverse buyer sentiment.
Select Clear to strive a special filter. For Entities, enter Hyundai
after which select Search. Choose the decision from the search outcomes and confirm from the transcript that the shopper was certainly calling about their Hyundai.
Search name transcripts
Transcript search is an experimental, non-compulsory, add-on function powered by Amazon Kendra.
Open the transcript net UI utilizing the URL proven as TranscriptionMediaSearchFinderURL
in the primary stack outputs. To discover a latest name, enter the search question buyer hit the wall
.
The outcomes present transcription extracts from related calls. Use the embedded audio participant to play the related part of the decision recording.
You possibly can increase Filter search outcomes to refine the search outcomes with further filters. Select Open Name Analytics to open the PCA name element web page for this name.
Question name analytics utilizing SQL
You possibly can combine PCA name analytics knowledge right into a reporting or BI instrument comparable to Amazon QuickSight through the use of Amazon Athena SQL queries. To strive it, open the Athena question editor. For Database, select pca.
Observe the desk parsedresults
. This desk incorporates all of the turn-by-turn transcriptions and evaluation for every name, utilizing nested buildings.
You too can overview flattened outcome units, that are less complicated to combine into your reporting or analytics software. Use the question editor to preview the info.
Processing circulation overview
How did PCA transcribe and analyze your cellphone name recordings? Let’s take a fast take a look at the way it works.
The next diagram exhibits the primary knowledge processing elements and the way they match collectively at a excessive degree.
Name recording audio information are uploaded to the S3 bucket and folder, recognized in the primary stack outputs as InputBucket
and InputBucketPrefix
, respectively. The pattern name recordings are mechanically uploaded since you set the parameter loadSampleAudioFiles
to true while you deployed PCA.
As every recording file is added to the enter bucket, an S3 Occasion Notification triggers a Lambda perform that initiates a workflow in Step Features to course of the file. The workflow orchestrates the steps to start out an Amazon Transcribe batch job and course of the outcomes by doing entity detection and extra preparation of the decision analytics outcomes. Processed outcomes are saved as JSON information in one other S3 bucket and folder, recognized in the primary stack outputs as OutputBucket
and OutputBucketPrefix
.
Because the Step Features workflow creates every JSON outcomes file within the output bucket, an S3 Occasion Notification triggers a Lambda perform, which masses chosen name metadata right into a DynamoDB desk.
The PCA UI net app queries the DynamoDB desk to retrieve the listing of processed calls to show on the house web page. The decision element web page reads further detailed transcription and analytics from the JSON outcomes file for the chosen name.
Amazon S3 Lifecycle insurance policies delete recordings and JSON information from each enter and output buckets after a configurable retention interval, outlined by the deployment parameter RetentionDays
. S3 Occasion Notifications and Lambda capabilities maintain the DynamoDB desk synchronized as information are each created and deleted.
When the EnableTranscriptKendraSearch
parameter is true
, the Step Features workflow additionally provides time markers and metadata attributes to the transcription, that are loaded into an Amazon Kendra index. The transcription search net software is used to look name transcriptions. For extra data on how this works, see Make your audio and video files searchable using Amazon Transcribe and Amazon Kendra.
Monitoring and troubleshooting
AWS CloudFormation studies deployment failures and causes on the stack Occasions tab. See Troubleshooting CloudFormation for assist with frequent deployment issues.
PCA offers runtime monitoring and logs for every part utilizing CloudWatch:
- Step Features workflow – On the Step Features console, open the workflow
PostCallAnalyticsWorkflow
. The Executions tab present the standing of every workflow run. Select any run to see particulars. Select CloudWatch Logs from the Execution occasion historical past to look at logs for any Lambda perform that was invoked by the workflow. - PCA server and UI Lambda capabilities – On the Lambda console, filter by
PostCallAnalytics
to see all of the PCA-related Lambda capabilities. Select your perform, and select the Monitor tab to see perform metrics. Select View logs in CloudWatch to examine perform logs.
Value evaluation
For pricing data for the primary providers utilized by PCA, see the next:
When transcription search is enabled, you incur an hourly value for the Amazon Kendra index: $1.125/hour for the Developer Version (first 750 hours are free), or $1.40/hour for the Enterprise Version (advisable for manufacturing workloads).
All different PCA prices are incurred primarily based on utilization, and are Free Tier eligible. After the Free Tier allowance is consumed, utilization prices add as much as about $0.15 for a 5-minute name recording.
To discover PCA prices for your self, use AWS Value Explorer or select Invoice Particulars on the AWS Billing Dashboard to see your month-to-date spend by service.
Combine together with your contact middle
You possibly can configure your contact middle to allow name recording. If potential, configure recordings for 2 channels (stereo), with buyer audio on one channel (for instance, channel 0) and the agent audio on the opposite channel (channel 1).
Through the AWS Command Line Interface (AWS CLI) or SDK, copy your contact middle recording information to the PCA enter bucket folder, recognized in the primary stack outputs as InputBucket
and InputBucketPrefix
. Alternatively, for those who already save your name recordings to Amazon S3, use deployment parameters InputBucketName
and InputBucketRawAudio
to configure PCA to make use of your present S3 bucket and prefix, so that you don’t have to repeat the information once more.
Customise your deployment
Use the next CloudFormation template parameters when creating or updating your stack to customise your PCA deployment:
- To allow or disable the non-compulsory (experimental) transcription search function, use
EnableTranscriptKendraSearch
. - To make use of your present S3 bucket for incoming name recordings, use
InputBucket
andInputBucketPrefix
. - To configure computerized deletion of recordings and name evaluation knowledge when utilizing auto-provisioned S3 enter and output buckets, use
RetentionDays
. - To detect name timestamp, agent title, or name identifier (GUID) from the recording file title, use
FilenameDatetimeRegex
,FilenameDatetimeFieldMap
,FilenameGUIDRegex
, andFilenameAgentRegex
. - To make use of the usual Amazon Transcribe API as an alternative of the default name analytics API, use TranscribeApiMode. PCA mechanically reverts to the usual mode API for audio recordings that aren’t appropriate with the decision analytics API (for instance, mono channel recordings). When utilizing the usual API some name analytics, metrics comparable to situation detection and speaker loudness aren’t out there.
- To set the listing of supported audio languages, use
TranscribeLanguages
. - To masks undesirable phrases, use
VocabFilterMode
and setVocabFilterName
to the title of a vocabulary filter that you just already created in Amazon Transcribe. See Vocabulary filtering for extra data. - To enhance transcription accuracy for technical and area particular acronyms and jargon, set
VocabularyName
to the title of a customized vocabulary that you just already created in Amazon Transcribe. See Customized vocabularies for extra data. - To configure PCA to make use of single-channel audio by default, and to determine audio system utilizing speaker diarizaton moderately than channel identification, use
SpeakerSeparationType
andMaxSpeakers
. The default is to make use of channel identification with stereo information utilizing Transcribe Name Analytics APIs to generate the richest analytics and most correct speaker labeling. - To redact PII from the transcriptions or from the audio, set
CallRedactionTranscript
orCallRedactionAudio
to true. See Redaction for extra data. - To customise entity detection utilizing Amazon Comprehend, or to offer your personal CSV file to outline entities, use the Entity detection parameters.
See the README on GitHub for extra particulars on configuration choices and operations for PCA.
PCA is an open-source undertaking. You possibly can fork the PCA GitHub repository, improve the code, and ship us pull requests so we are able to incorporate and share your enhancements!
Clear up
Once you’re completed experimenting with this resolution, clear up your assets by opening the AWS CloudFormation console and deleting the PostCallAnalytics
stacks that you just deployed. This deletes assets that you just created by deploying the answer. S3 buckets containing your audio recordings and analytics, and CloudWatch log teams are retained after the stack is deleted to keep away from deleting your knowledge.
Dwell Name Analytics: Companion resolution
Our companion resolution, Dwell Name Analytics (LCA), provides actual time-transcription and analytics capabilities through the use of the Amazon Transcribe and Amazon Comprehend real-time APIs. In contrast to PCA, which transcribes and analyzes recorded audio after the decision has ended, LCA transcribes and analyzes your calls as they’re occurring and offers real-time updates to supervisors and brokers. You possibly can configure LCA to retailer name recordings to the PCA’s ingestion S3 bucket, and use the 2 options collectively to get the perfect of each worlds. See Live call analytics for your contact center with Amazon language AI services for extra data.
Conclusion
The Publish Name Analytics resolution provides a scalable, cost-effective method to offer name analytics with options to assist enhance your callers’ expertise. It makes use of Amazon ML providers like Transcribe Name Analytics and Amazon Comprehend to transcribe and extract wealthy insights out of your buyer conversations.
The pattern PCA software is supplied as open supply—use it as a place to begin in your personal resolution, and assist us make it higher by contributing again fixes and options by way of GitHub pull requests. For skilled help, AWS Skilled Providers and different AWS Companions are right here to assist.
We’d love to listen to from you. Tell us what you assume within the feedback part, or use the problems discussion board within the PCA GitHub repository.
In regards to the Authors
Dr. Andrew Kane is an AWS Principal WW Tech Lead (AI Language Providers) primarily based out of London. He focuses on the AWS Language and Imaginative and prescient AI providers, serving to our clients architect a number of AI providers right into a single use-case pushed resolution. Earlier than becoming a member of AWS at first of 2015, Andrew spent twenty years working within the fields of sign processing, monetary funds programs, weapons monitoring, and editorial and publishing programs. He’s a eager karate fanatic (only one belt away from Black Belt) and can be an avid home-brewer, utilizing automated brewing {hardware} and different IoT sensors.
Bob Strahan is a Principal Options Architect within the AWS Language AI Providers workforce.
Connor Kirkpatrick is an AWS Options Engineer primarily based within the UK. Connor works with the AWS Answer Architects to create standardised instruments, code samples, demonstrations, and quickstarts. He’s an enthusiastic rower, wobbly bicycle owner, and occasional baker.
Franco Rezabek is an AWS Options Engineer primarily based in London, UK. Franco works with AWS Answer Architects to create standardized instruments, code samples, demonstrations, and fast begins.
Steve Engledow is a Options Engineer working with inner and exterior AWS clients to construct reusable options to frequent issues.