Genesys Cloud - Main

Leor,

Great to hear!  Question for you.  Would this eventually be able to consume grammars like the Nuance libraries or open source ones like Kaldi?

Hey Robert!

Thanks! On initial release, it will support terms the same way it does today through the API. The next step will likely be capitalization support and special characters. Grammars aren't likely to be supported as the language models work a bit differently. The "Sounds like" feature helps build the equivalent.

Curious - are you seeing much usage of grammars in newer transcription engines? 

Thanks again,

Leor

Leor,

Great to hear!  Question for you.  Would this eventually be able to consume grammars like the Nuance libraries or open source ones like Kaldi?

Hey Leor,

From what we see are the following:

1. Insurance and heath benefit terms
2. City/state slots
3. Automotive parts
4. Addresses
5. Doctor names
6. Physician specialties

Hey Robert!

Thanks! On initial release, it will support terms the same way it does today through the API. The next step will likely be capitalization support and special characters. Grammars aren't likely to be supported as the language models work a bit differently. The "Sounds like" feature helps build the equivalent.

Curious - are you seeing much usage of grammars in newer transcription engines? 

Thanks again,

Leor

Leor,

Great to hear!  Question for you.  Would this eventually be able to consume grammars like the Nuance libraries or open source ones like Kaldi?

Hey Leor,

From what we see are the following:

1. Insurance and heath benefit terms
2. City/state slots
3. Automotive parts
4. Addresses
5. Doctor names
6. Physician specialties

Hey Robert!

Thanks! On initial release, it will support terms the same way it does today through the API. The next step will likely be capitalization support and special characters. Grammars aren't likely to be supported as the language models work a bit differently. The "Sounds like" feature helps build the equivalent.

Curious - are you seeing much usage of grammars in newer transcription engines? 

Thanks again,

Leor

Leor,

Great to hear!  Question for you.  Would this eventually be able to consume grammars like the Nuance libraries or open source ones like Kaldi?

Hi Leor,

This is amazing!!

The launch of this functionality is still May 20, if not, when will the launch of this one be?

Thanks,
Francisco Alvarez

Yes I don't see this on the release notes for this week.  What's our new ETA please Leor?  Looking forward to this one, but even more looking forward to the later recording example support.  We have similar use cases as what Robert's mentioned in the medical/healthcare industry.

Hi Leor,

This is amazing!!

The launch of this functionality is still May 20, if not, when will the launch of this one be?

Thanks,
Francisco Alvarez

I still can't see this feature even though the release notes say it was released yesterday can you please confirm if it did get pushed out? Very excited about this feature. 

Hi Tanya! We're slowly rolling this out across regions. It should be fully enabled in all regions by tomorrow evening Eastern Time. 

I still can't see this feature even though the release notes say it was released yesterday can you please confirm if it did get pushed out? Very excited about this feature. 

Hi Leor. I have been working with a client to improve recognition of some of the Australian council areas - we tend to roll words together. From tests so far it definitely has helped from recognising one specific area from about 10% to 50%. However I have found that using the new GUI you cannot set the boost value and when pulling data back through the API it defaults to "2". I have now increased the boost to "10" via the API and recognition is much higher. My question is, will the boost setting be added to the GUI? Is there a reason why it was not included in the first instance?

Hi Tanya! We're slowly rolling this out across regions. It should be fully enabled in all regions by tomorrow evening Eastern Time. 

I still can't see this feature even though the release notes say it was released yesterday can you please confirm if it did get pushed out? Very excited about this feature. 

Hi Malcolm,

Thanks so much for the feedback. Very happy to hear the initial successes! 

We're looking at adding the boost functionality into the UI as well - part of the conversation was trading off simplicity in the UI vs increased functionality. The default of 2 should work for most terms and being able to set everything to 10 could end up leading to a decrease in accuracy in some instances.

Will update as we look to add more functionality to this feature.

Leor

Hi Leor. I have been working with a client to improve recognition of some of the Australian council areas - we tend to roll words together. From tests so far it definitely has helped from recognising one specific area from about 10% to 50%. However I have found that using the new GUI you cannot set the boost value and when pulling data back through the API it defaults to "2". I have now increased the boost to "10" via the API and recognition is much higher. My question is, will the boost setting be added to the GUI? Is there a reason why it was not included in the first instance?

Hi Tanya! We're slowly rolling this out across regions. It should be fully enabled in all regions by tomorrow evening Eastern Time. 

I still can't see this feature even though the release notes say it was released yesterday can you please confirm if it did get pushed out? Very excited about this feature. 

Hi Leor,

I just started working with this feature last week around tuning for some medical terminology, specifically to see how I can add medical terms so that Knowledge Articles can be surfaced via Agent Assist, and those same terms can show up correctly in the auto-generated summary. A few questions came up:

1. When I add a term, and associated phrases, how long does it take for the transcription engine to take into account those new words? Is it immediate as the next call?
2. What is your guidance on tuning this experience? Say I am introducing a new medical product with a unique name. How would you go about handling that term? Would you add the term, some sounds like phrases and some examples, and keep it on the default "2" boost setting? Then, as time goes on, slowly increase the boost setting to 3, then to 4, etc. to see how it affects the transcription accuracy? Curious to know your thoughts on this.  The only visibility I can think of that an Admin has is relying on agents feedback on accuracy, or just going to look at conversation transcripts 1 by 1, which could be a large manual effort.

A couple of other thoughts I had:

1. For any unique terms that are defined in Dictionary Management, I think it would be really interesting to be able to expose some ability for agents to select from those terms to "correct" the summary. So, in this scenario the call is completed and the summary is generated. If the transcription engine transcribes a word -- say "particle", and the dictionary Term is Partilique --, and the Admin has defined particle as a "sounds like" phrase.. It would be interesting to provide some sort of indiciation to the agent that "hey, I found another possible transcription option.." and offer of "Partililque" as the transcription correction, where the agent could select that as the proper transcription term.
2. Even cooler, would be the ability to offer that sort of suggestion in real-time.. so that the agent could make that correct, which could then in downstream affect knowledge surfacing, etc.

Anyways, the maintenance and tuning of this is really interesting to me. Curious to know what you're thinking up over there and what's on the vision board/roadmap. 

Thanks,

Peter

Hi Malcolm,

Thanks so much for the feedback. Very happy to hear the initial successes! 

We're looking at adding the boost functionality into the UI as well - part of the conversation was trading off simplicity in the UI vs increased functionality. The default of 2 should work for most terms and being able to set everything to 10 could end up leading to a decrease in accuracy in some instances.

Will update as we look to add more functionality to this feature.

Leor

Hi Leor. I have been working with a client to improve recognition of some of the Australian council areas - we tend to roll words together. From tests so far it definitely has helped from recognising one specific area from about 10% to 50%. However I have found that using the new GUI you cannot set the boost value and when pulling data back through the API it defaults to "2". I have now increased the boost to "10" via the API and recognition is much higher. My question is, will the boost setting be added to the GUI? Is there a reason why it was not included in the first instance?

Hi Tanya! We're slowly rolling this out across regions. It should be fully enabled in all regions by tomorrow evening Eastern Time. 

I still can't see this feature even though the release notes say it was released yesterday can you please confirm if it did get pushed out? Very excited about this feature. 

Hi Peter, thanks for the questions.

1. Typically, the language model is updated within 15 minutes. Often, less.
2. Yes, keep at the default (which is the only option in the UI) and increase by a single digit until the desired results. Over-boosting that lead to over-identifying the term. 

We are looking at building in feedback mechanisms to the UI and elsewhere to help make it easier to report issues. Very grateful for these ideas! Will share more as we get closer to updating the roadmap.

Leor

Hi Leor,

I just started working with this feature last week around tuning for some medical terminology, specifically to see how I can add medical terms so that Knowledge Articles can be surfaced via Agent Assist, and those same terms can show up correctly in the auto-generated summary. A few questions came up:

1. When I add a term, and associated phrases, how long does it take for the transcription engine to take into account those new words? Is it immediate as the next call?
2. What is your guidance on tuning this experience? Say I am introducing a new medical product with a unique name. How would you go about handling that term? Would you add the term, some sounds like phrases and some examples, and keep it on the default "2" boost setting? Then, as time goes on, slowly increase the boost setting to 3, then to 4, etc. to see how it affects the transcription accuracy? Curious to know your thoughts on this.  The only visibility I can think of that an Admin has is relying on agents feedback on accuracy, or just going to look at conversation transcripts 1 by 1, which could be a large manual effort.

A couple of other thoughts I had:

1. For any unique terms that are defined in Dictionary Management, I think it would be really interesting to be able to expose some ability for agents to select from those terms to "correct" the summary. So, in this scenario the call is completed and the summary is generated. If the transcription engine transcribes a word -- say "particle", and the dictionary Term is Partilique --, and the Admin has defined particle as a "sounds like" phrase.. It would be interesting to provide some sort of indiciation to the agent that "hey, I found another possible transcription option.." and offer of "Partililque" as the transcription correction, where the agent could select that as the proper transcription term.
2. Even cooler, would be the ability to offer that sort of suggestion in real-time.. so that the agent could make that correct, which could then in downstream affect knowledge surfacing, etc.

Anyways, the maintenance and tuning of this is really interesting to me. Curious to know what you're thinking up over there and what's on the vision board/roadmap. 

Thanks,

Peter

Hi Malcolm,

Thanks so much for the feedback. Very happy to hear the initial successes! 

We're looking at adding the boost functionality into the UI as well - part of the conversation was trading off simplicity in the UI vs increased functionality. The default of 2 should work for most terms and being able to set everything to 10 could end up leading to a decrease in accuracy in some instances.

Will update as we look to add more functionality to this feature.

Leor

Hi Leor. I have been working with a client to improve recognition of some of the Australian council areas - we tend to roll words together. From tests so far it definitely has helped from recognising one specific area from about 10% to 50%. However I have found that using the new GUI you cannot set the boost value and when pulling data back through the API it defaults to "2". I have now increased the boost to "10" via the API and recognition is much higher. My question is, will the boost setting be added to the GUI? Is there a reason why it was not included in the first instance?

Hi Tanya! We're slowly rolling this out across regions. It should be fully enabled in all regions by tomorrow evening Eastern Time. 

I still can't see this feature even though the release notes say it was released yesterday can you please confirm if it did get pushed out? Very excited about this feature. 

I may have missed this when the original API was launched, but will this dictionary management also apply to the accuracy of bot flows using the native Genesys STT (ask for intent and ask for slot)?  My company's product is way too often "heard" by the bot as "goat you connect" and other awful misunderstandings, instead of "GoTo Connect".

Currently, voice bots use an older technology that isn't the same as the existing voice transcription engine. We are looking at bringing that engine to voice bots, in which case Dictionary Management would apply.

Currently, voice bots use an older technology that isn't the same as the existing voice transcription engine. We are looking at bringing that engine to voice bots, in which case Dictionary Management would apply.

I may have missed this when the original API was launched, but will this dictionary management also apply to the accuracy of bot flows using the native Genesys STT (ask for intent and ask for slot)?  My company's product is way too often "heard" by the bot as "goat you connect" and other awful misunderstandings, instead of "GoTo Connect".

It was touched on earlier in the thread, but want to confirm - do Dictionary Management terms apply to Agent Assist/CoPilot transcription and knowledge article surfacing?

Thanks,

Brady

------------------------------
Brady
Seattle, USA
------------------------------

Hi Brady, 

Thanks for the note. If the user has GC3 or WEM Add On today, what's added to Dictionary Management will affect the transcription that is consumed by Agent Assist and Copilot if Native Voice Transcription is used. We hope to be able to provide the Dictionary Management soon to those who might not have the GC3 or WEM Add On licenses soon. @Amanda Halpin for visibility.

Leor

It was touched on earlier in the thread, but want to confirm - do Dictionary Management terms apply to Agent Assist/CoPilot transcription and knowledge article surfacing?

Thanks,

Brady

------------------------------
Brady
Seattle, USA

Leor, 

I think what people are really asking is how do we make the transcription/ASR/Topic Spotting the best it can be.  Other companies offer pre-trained dictionaries, the ability to feed in lots of sentences and/or words, and the ability to train dictionaries by using audio or spoken input.  The 500 entry limit is very limiting, especially in the medical field.  If you were to ask me for my top 10, here is what I would suggest:

1. Empathy Feedback like we have for Sentiment
2. Ability to train more than 500 dictionary words in bulk without taking 30-50 clicks per word
3. Allow dictionaries to be trained on actual voice input of the word instead of just phonetics
4. Allow AI Generation of phonetic pronunciation like Bing replied to this for Genesys:  juh-NEH-sis
5. Genesys should publish pre-trained dictionaries for Medical, Insurance, Financial, and other industries
6. Correction mechanism in transcript to train model when inaccurate
7. Transcription Feedback loop for LM on positive or negative input from agents and/or QM Managers
8. Speaker-based output of transcription via the API
9. Dialect and accent detection and adjustment after the fact
10. Translation of transcriptions (wait - that one is coming!)

Hi Brady, 

Thanks for the note. If the user has GC3 or WEM Add On today, what's added to Dictionary Management will affect the transcription that is consumed by Agent Assist and Copilot if Native Voice Transcription is used. We hope to be able to provide the Dictionary Management soon to those who might not have the GC3 or WEM Add On licenses soon. @Amanda Halpin for visibility.

Leor

It was touched on earlier in the thread, but want to confirm - do Dictionary Management terms apply to Agent Assist/CoPilot transcription and knowledge article surfacing?

Thanks,

Brady

------------------------------
Brady
Seattle, USA

Leor, 

I think what people are really asking is how do we make the transcription/ASR/Topic Spotting the best it can be.  Other companies offer pre-trained dictionaries, the ability to feed in lots of sentences and/or words, and the ability to train dictionaries by using audio or spoken input.  The 500 entry limit is very limiting, especially in the medical field.  If you were to ask me for my top 10, here is what I would suggest:

1. Empathy Feedback like we have for Sentiment
2. Ability to train more than 500 dictionary words in bulk without taking 30-50 clicks per word
3. Allow dictionaries to be trained on actual voice input of the word instead of just phonetics
4. Allow AI Generation of phonetic pronunciation like Bing replied to this for Genesys:  juh-NEH-sis
5. Genesys should publish pre-trained dictionaries for Medical, Insurance, Financial, and other industries
6. Correction mechanism in transcript to train model when inaccurate
7. Transcription Feedback loop for LM on positive or negative input from agents and/or QM Managers
8. Speaker-based output of transcription via the API
9. Dialect and accent detection and adjustment after the fact
10. Translation of transcriptions (wait - that one is coming!)

Hi Brady, 

Thanks for the note. If the user has GC3 or WEM Add On today, what's added to Dictionary Management will affect the transcription that is consumed by Agent Assist and Copilot if Native Voice Transcription is used. We hope to be able to provide the Dictionary Management soon to those who might not have the GC3 or WEM Add On licenses soon. @Amanda Halpin for visibility.

Leor

It was touched on earlier in the thread, but want to confirm - do Dictionary Management terms apply to Agent Assist/CoPilot transcription and knowledge article surfacing?

Thanks,

Brady

------------------------------
Brady
Seattle, USA

Hi Brady, 

Thanks for the note. If the user has GC3 or WEM Add On today, what's added to Dictionary Management will affect the transcription that is consumed by Agent Assist and Copilot if Native Voice Transcription is used. We hope to be able to provide the Dictionary Management soon to those who might not have the GC3 or WEM Add On licenses soon. @Amanda Halpin for visibility.

Leor

------------------------------
Leor Grebler, Principal Product Manager - Voice Transcription
leor.grebler@genesys.com
Genesys - Employees
------------------------------


Original Message:
Sent: 09-23-2024 10:57
From: Brady Speed
Subject: Upcoming New Feature - Dictionary Management in UI

It was touched on earlier in the thread, but want to confirm - do Dictionary Management terms apply to Agent Assist/CoPilot transcription and knowledge article surfacing?

Thanks,

Brady

------------------------------
Brady
Seattle, USA

Thanks for the info!

Follow up question - how is "confidence" derived as seen in the transcript URL result for the below API?

 api/v2/speechandtextanalytics/conversations/{conversationId}/communications/{communicationId}/transcripturl

"words": [            {              "word": "and",              "confidence": 0.971,              "startTimeMs": 1726155205633,              "duration": {                "milliseconds": 40              }            }

------------------------------
Brady
Seattle, USA
------------------------------


Original Message:
Sent: 09-23-2024 16:19
From: Leor Grebler
Subject: Upcoming New Feature - Dictionary Management in UI

Hi Brady, 

Thanks for the note. If the user has GC3 or WEM Add On today, what's added to Dictionary Management will affect the transcription that is consumed by Agent Assist and Copilot if Native Voice Transcription is used. We hope to be able to provide the Dictionary Management soon to those who might not have the GC3 or WEM Add On licenses soon. @Amanda Halpin for visibility.

Leor

------------------------------
Leor Grebler, Principal Product Manager - Voice Transcription
leor.grebler@genesys.com
Genesys - Employees


Original Message:
Sent: 09-23-2024 10:57
From: Brady Speed
Subject: Upcoming New Feature - Dictionary Management in UI

It was touched on earlier in the thread, but want to confirm - do Dictionary Management terms apply to Agent Assist/CoPilot transcription and knowledge article surfacing?

Thanks,

Brady

------------------------------
Brady
Seattle, USA