[ad_1]
|
In 2017, we launched Amazon Transcribe, an automated speech recognition (ASR) service that makes it simple so as to add speech-to-text capabilities to any software. At present, I’m very completely happy to announce the provision of Amazon Transcribe Name Analytics, a brand new function that allows you to simply extract worthwhile insights from buyer conversations with a single API name.
Every dialogue with potential or current clients is a chance to find out about their wants and expectations. For instance, it’s vital for customer support groups to determine the primary the reason why clients are calling them, and measure buyer satisfaction throughout these calls. Likewise, salespeople attempt to gauge buyer curiosity, and their response to a specific gross sales pitch.
Thus, many shoppers and companions want to add name analytics capabilities in numerous functions, no matter their contact heart supplier. They typically want to investigate greater than cellphone calls, for instance web-based audio and video calls. Up to now, they’ve usually achieved this by stitching AI companies and devoted ML fashions collectively, they usually’ve requested us for an easier answer.
We set to work and constructed Amazon Transcribe Name Analytics, a brand new addition to Transcribe and a key enhancement to AWS Contact Middle Intelligence. In the event you can’t wait to strive it, be happy to leap now to the AWS console. In the event you’d prefer to study extra, learn on!
Introducing Amazon Transcribe Name Analytics
Based mostly on ASR applied in Transcribe, Transcribe Name Analytics provides pure language processing (NLP) capabilities particularly skilled on buyer calls, and optimized to supply extremely correct name transcripts and actionable insights. With a easy API name, builders can now simply add name analytics to any software, and extract buyer insights from conversations with out having to construct AI pipelines and prepare customized ML fashions.
Key options of Transcribe Name Analytics embody:
- Timestamped turn-by-turn name transcription in 21 languages.
- Situation detection, which picks up the shortest set of contiguous phrases in a dialog flip that represents the explanation why the shopper is asking. This works out of the field with none configuration or coaching.
- Name categorization primarily based on conversational traits:
- Matching particular phrases and phrases,
- Detecting non-talk time,
- Detecting interruptions,
- Analyzing sentiment for the shopper and the agent.
- Name traits resembling:
- How shortly and loudly a buyer or agent are talking,
- Detecting non-talk time,
- Detecting interruptions.
- Redaction of delicate information from the textual content transcript and the corresponding audio file.
For instance, you’ll be able to create guidelines to flag calls the place clients interrupt the agent, exhibit detrimental sentiment, and say “I wish to communicate with the supervisor”. These calls definitely didn’t go properly, and are price analyzing intimately! It’s also possible to search for calls the place brokers don’t use pre-defined greetings (“Welcome to ACME Help, how can I make it easier to as we speak?”) inside the first 15 seconds, to measure script compliance and assist supervisors determine agent teaching alternatives. One other in style situation is to create guidelines that flag mentions of your particular services (“Your ACME Turbo 2000 vacuum cleaner isn’t working prefer it ought to”), with a purpose to choose up any rising tendencies you’d want to pay attention to.
Final however not least, you’ll be able to additional course of the textual content transcript with different AI companies resembling Amazon Translate, or with customized NLP fashions constructed with Amazon SageMaker.
Now, let’s do a fast demo.
Extracting Insights with Amazon Transcribe Name Analytics
Right here’s a fictitious help name, the place a woman calls her financial institution to report that she’s misplaced her credit score and debit playing cards. The sound file is a stereo WAV file (16-bit, 8KHz).
Transcribe Name Analytics requires that the agent and the shopper are recorded in their very own channel. We’ll additionally want to inform which is the agent channel. In a stereo file, the left channel is often the primary channel (channel #Zero), and the correct channel is the second (channel #1). That is the case for this name.
In the event you’re undecided which is which, you’ll be able to simply use the versatile ffmpeg open supply software to extract every channel to a separate audio file.
$ ffmpeg -i demo-call.wav -map_channel Zero.Zero.Zero channel0.wav -map_channel Zero.Zero.1 channel1.wav
You need to use the identical method to extract audio channels from different file sorts, resembling video recordsdata, and recombine them to a stereo audio file. You’ll discover extra info within the ffmpeg documentation.
Now that I’m certain that the agent is in channel #1, I take advantage of the AWS CLI to add the audio file to an S3 bucket.
$ aws s3 cp launch-call.wav s3://jsimon-transcribe-useast1/demo-call.wav --region us-east-1
Opening the Transcribe Name Analytics console, I see that decision class templates can be found.
I resolve to create one for supervisor escalations. Then, with a few clicks, I create a customized name class named welcome-message, to test if the agent begins the decision with an applicable welcome. I may add a number of phrases to test for if wanted. We suggest that you just use quick sentences to attenuate the prospect of filler phrases popping up (‘hmm’, ‘err’, and so forth).
Then, I create a name analytics job utilizing the final mannequin obtainable in Transcribe. I additionally allow automated language detection.
Then, I outline the placement of the audio file in S3, flagging channel #1 because the agent channel.
I resolve to retailer the transcript within the default S3 bucket created by Transcribe in my account. I may additionally use my very own bucket if wanted. Then, I choose an AWS Id and Entry Administration (IAM) function with enough permissions, and I launch the job.
A minute later or so, the job is full. The console comprises a preview of the textual content transcript, in addition to a hyperlink to the total JSON transcript.
Because the agent used the right welcome sentence within the first 15 seconds, the decision is tagged with the class I created earlier.
Downloading the JSON transcript, every sentence within the dialog is enriched with metadata on per-word loudness, measured on a Zero-100 vary with 100 being extraordinarily loud. Right here’s the primary sentence:
"BeginOffsetMillis":440,"EndOffsetMillis":4960,
"Sentiment":"NEUTRAL",
"ParticipantRole":"AGENT",
"LoudnessScores":[78.68,80.4,81.91,78.95,82.34],
"Content material":"Whats up and thanks for calling the financial institution. That is Ashley talking, how might I make it easier to as we speak?"
Trying on the subsequent sentence, I see that Transcribe Name Analytics mechanically detected what the shopper situation is. The corresponding textual content is in daring:
"Content material": "Hello um uh you simply must cancel my card. Um I've a debit card and a bank card.","IssuesDetected":[. . .
On the finish of the transcript, I see international name statistics (length, speak time, phrases per minute, matched classes). Transcribe additionally offers me total sentiment info, meaured from -5 (extraordinarily detrimental) to +5 (extraordinarily constructive). I additionally get a a breakdown in 4 quarters.
"Sentiment":{"OverallSentiment":"AGENT":2.6,"CUSTOMER":Zero.2,
"SentimentByPeriod":"QUARTER":
}
We will see that the shopper began the decision with detrimental sentiment, transferring shortly to impartial sentiment, and ending the decision with constructive sentiment. It is a good signal that the decision was dealt with satisfactorily, and that the shopper downside was solved.
In the event you’d prefer to convert the transcript to a Phrase doc with extra visualizations, my colleague Andrew Kane has constructed a pleasant software and made it obtainable on Github. Right here’s a pattern report produced by his software.
AWS Clients and Companions Are Utilizing Amazon Transcribe Name Analytics
Ben Rigby, the SVP, World Head of Product & Engineering, Synthetic Intelligence, Automation, and Workforce at Talkdesk informed us, “Our clients are processing thousands and thousands of customer support calls of their contact facilities a 12 months and have a crucial must extract actionable dialog insights to make sure constructive enterprise outcomes. As an AWS Contact Middle Intelligence companion, we additional enhanced our name transcription capabilities with Amazon Transcribe. With the launch of Amazon Transcribe Name Analytics, we’re excited so as to add much more AI capabilities to our Speech Analytics and QM Help merchandise. These deeper insights can present brokers and supervisors with the info they should enhance the pace and high quality of their customer support whereas boosting workforce productiveness.”
Praphul Kumar, the Chief Product Officer of SuccessKPI provides, “Amazon Transcribe Name Analytics API permits us so as to add ML-based capabilities to our platform sooner and at a decrease value. This new API removes the necessity to combine a number of AI companies collectively and develop customized machine studying fashions in sure areas. With Transcribe Name Analytics, we will present dialog insights resembling sentiment, non-talk time, and name classes to gauge agent efficiency. This helps to drive higher name outcomes, cut back agent turnover, uncover agent teaching alternatives, and measure name script compliance. Combining AWS companies into SuccessKPI’s expertise analytics platform was a no brainer. We’re wanting ahead to bringing this worthwhile functionality into the palms of enormous enterprises and authorities businesses.”
Getting Began
A single API name is all it takes to extract wealthy insights out of your buyer conversations. You can begin utilizing Amazon Transcribe Name Analytics as we speak within the following areas:
- US West (Oregon), US East (N. Virginia),
- Canada (Central),
- Europe (London), Europe (Frankfurt),
- Asia Pacific (Mumbai), Asia Pacific (Singapore), Asia Pacific (Seoul), Asia Pacific (Tokyo), Asia Pacific (Sydney).
Please give this new function a strive within the AWS console, and tell us what you assume. We at all times look ahead to your suggestions! You may ship it by way of your regular AWS Help contacts or put up it on the AWS Discussion board for Amazon Transcribe.
One very last thing: when you’re in search of a straightforward to make use of omnichannel cloud contact heart, you need to undoubtedly check out Amazon Join and its ML powered analytics, Contact Lens.
– Julien
[ad_2]
Source link
