Speech and interaction analytics

See also

Sources

1. 1
   Amazon Connect voice conversational analytics requires call recording of both agent and customer to be enabled in the Set recording and analytics behavior block.

   Enable conversational analytics in Connect Customer conversational analytics · section 'Enable call recording and speech analytics', step 2 · Checked 2026-10-01
2. 2
   Amazon Connect email analytics has no real-time versus post-contact distinction and produces no sentiment scores; it provides categorization, redaction and contact summaries.

   Analyze conversations using conversational analytics in Connect Customer conversational analytics · 'Email analytics support' list and 'Sample Contact details page for email analytics' · Checked 2026-10-01
3. 3
   Amazon Connect AI-feature language entries marked with an asterisk are not available in instances in the Africa (Cape Town) and AWS GovCloud (US-West) Regions.

   Languages supported by Connect Customer features · section 'AI features', footnote marked * · Checked 2026-10-01
4. 4
   Amazon Connect Natural Language Semantic Match (generative AI) rule conditions cannot be used for real-time analysis and require the 'Rules - Generative AI' permission.

   Automatically categorize contacts by matching conversations with natural language statements, or specific words and phrases · section 'Step 1: Define conditions', Note under Natural Language - Semantic Match · Checked 2026-10-01
5. 5
   Amazon Connect writes the original (unredacted) analyzed file only when 'Get redacted and original transcripts with redacted audio' is chosen, and for voice it is the only complete record of the conversation.

   Use sensitive data redaction to protect customer privacy using conversational analytics · section 'About redacted files', original analyzed file bullet and Important box · Checked 2026-10-01
6. 6
   AWS states that the Amazon Connect post-call analytics option, which analyzes the recording after the conversation and After Contact Work complete, provides the best transcription accuracy.

   Enable conversational analytics in Connect Customer conversational analytics · section 'Enable call recording and speech analytics', step 4a · Checked 2026-10-01
7. 7
   Some Amazon Connect languages, for example Hungarian (hu_HU), Turkish (tr_TR) and Tamil (ta_IN), are supported for post-call analytics but not for real-time call analytics.

   Languages supported by Connect Customer features · section 'AI features' table, rows Hungarian (Hungary), Turkish (Turkey), Tamil (India) · Checked 2026-10-01
8. 8
   Amazon Connect conversational analytics offers both real-time call analytics (while the call is in progress) and post-call analytics for voice contacts.

   Analyze conversations using conversational analytics in Connect Customer conversational analytics · intro, 'Speech analytics support' list · Checked 2026-10-01
9. 9
   Choosing the Amazon Connect real-time analytics option gives real-time insights during the call and also post-call analytics after the conversation and After Contact Work end.

   Enable conversational analytics in Connect Customer conversational analytics · section 'Enable call recording and speech analytics', step 4b · Checked 2026-10-01
10. 10
    For Amazon Connect voice contacts, sensitive data redaction is applied after the call disconnects.

    Use sensitive data redaction to protect customer privacy using conversational analytics · intro paragraph before the Important box · Checked 2026-10-01
11. 11
    In Amazon Connect redacted audio files, sensitive data is replaced with silence, and these silences are not flagged as non-talk time.

    Use sensitive data redaction to protect customer privacy using conversational analytics · section 'About redacted files', redacted audio file bullet · Checked 2026-10-01
12. 12
    In Amazon Connect, Japanese (ja_JP) and Korean (ko_KR) are supported for post-call and real-time call analytics and sentiment but not for redaction.

    Languages supported by Connect Customer features · section 'AI features' table, rows Japanese (Japan) and Korean (South Korea), Redaction column · Checked 2026-10-01
13. 13
    AWS states that Amazon Connect redaction may not identify and remove all instances of sensitive data because of the predictive nature of machine learning, and recommends reviewing redacted output.

    Use sensitive data redaction to protect customer privacy using conversational analytics · Important box under the intro · Checked 2026-10-01
14. 14
    AWS states Amazon Connect redaction does not meet HIPAA de-identification requirements and recommends continuing to treat redacted output as protected health information.

    Use sensitive data redaction to protect customer privacy using conversational analytics · Important box under the intro, second paragraph · Checked 2026-10-01
15. 15
    Amazon Connect redaction is supported only for post-call analytics and chat analytics and is not supported for real-time call analytics.

    Languages supported by Connect Customer features · section 'AI features', footnote marked ** · Checked 2026-10-01
16. 16
    Amazon Connect conversational analytics rule actions include generating an EventBridge event, creating a task, creating a case, sending email notifications and submitting an automated evaluation.

    Automatically categorize contacts by matching conversations with natural language statements, or specific words and phrases · section 'Step 2: Define actions' · Checked 2026-10-01
17. 17
    Amazon Connect categorization rules can add conditions on queue, contact attribute values and sentiment score in addition to word, phrase or natural-language matches.

    Automatically categorize contacts by matching conversations with natural language statements, or specific words and phrases · section 'Step 1: Define conditions', step 9 · Checked 2026-10-01
18. 18
    Amazon Connect conversational analytics rules support four condition match types: Words or Phrases Exact Match, Words or Phrases Pattern Match, Natural Language Semantic Match and Words or Phrases Semantic Match.

    Automatically categorize contacts by matching conversations with natural language statements, or specific words and phrases · section 'Step 1: Define conditions', step 6 · Checked 2026-10-01
19. 19
    Amazon Connect conversational analytics rules apply only to contacts that occur after the rule is added; they cannot be applied to past, stored conversations.

    Automatically categorize contacts by matching conversations with natural language statements, or specific words and phrases · section 'Step 3: Review and save', step 2 · Checked 2026-10-01
20. 20
    In Amazon Connect, sentiment analysis is on by default when a sentiment-supported language is chosen and speech, chat or email analytics is enabled, and it can be deselected in the flow block.

    Enable conversational analytics in Connect Customer conversational analytics · sections 'Important things to know' and 'Disable sentiment analysis' · Checked 2026-10-01
21. 21
    The Amazon Connect overall sentiment score for a contact is the average of the scores assigned to each portion of the call.

    Investigate sentiment scores during contact conversations using conversational analytics · section 'How sentiment scores are determined', final sentence · Checked 2026-10-01
22. 22
    Amazon Connect classifies each speaker turn as positive, negative or neutral and assigns a sentiment score from -5 to +5 for each period of the call.

    Investigate sentiment scores during contact conversations using conversational analytics · section 'How sentiment scores are determined' · Checked 2026-10-01
23. 23
    An Amazon Connect transfer creates a second contact ID, so analytics continues after an agent or queue transfer only if the transfer flow has another Set recording and analytics behavior block with analytics enabled; queue-to-queue transfers copy the configuration.

    Enable conversational analytics in Connect Customer conversational analytics · section 'Important things to know', first bullet and Note · Checked 2026-10-01
24. 24
    Amazon Connect conversational analytics supports calls with up to two participants; with more parties transcription and analytics quality can degrade, and AWS recommends disabling analytics for such calls.

    Enable conversational analytics in Connect Customer conversational analytics · section 'Multi-party calls and conversational analytics' · Checked 2026-10-01
25. 25
    Genesys Cloud AudioHook Monitor, a premium AppFoundry application, streams conversation audio to third-party vendors for uses such as transcription, voice biometrics, recording and agent assist.

    About AudioHook Monitor · article body, overview paragraph · Checked 2026-10-01
26. 26
    Genesys states AudioHook Monitor cannot be used to stream audio during secure flows for PCI use cases.

    About AudioHook Monitor · article body, PCI DSS note · Checked 2026-10-01
27. 27
    Genesys Cloud dictionary management lets administrators add organization-specific terms such as brand names and acronyms to improve voice transcription accuracy.

    About voice transcription · section 'Improving transcription accuracy' · Checked 2026-10-01
28. 28
    Genesys Cloud offers native voice transcription and Extended Voice Transcription Services (EVTS), which supports third-party transcription providers including Microsoft Azure, Amazon and Google.

    About voice transcription · article body, transcription engine options · Checked 2026-10-01
29. 29
    Genesys Cloud speech and text analytics runs immediately after an interaction ends; voice transcripts with lower latency can be obtained by subscribing through the Notifications API.

    Speech and text analytics overview · article body, transcript availability paragraph · Checked 2026-10-01
30. 30
    Genesys Cloud programs can control whether customer sentiment analysis and agent empathy analysis run for interactions associated with the program.

    About programs, topics, and phrases · article body, programs paragraph · Checked 2026-10-01
31. 31
    In Genesys Cloud, a program is a package of topics, a topic is a collection of phrases that indicate a business intent, and a phrase is one way a topic can be expressed.

    About programs, topics, and phrases · article body, definitions of programs, topics and phrases · Checked 2026-10-01
32. 32
    Genesys states that low-quality voice transcriptions can cause missing or mislabeled positive or negative sentiment markers.

    Understand sentiment analysis · article body, limitation note on transcription quality · Checked 2026-10-01
33. 33
    Genesys Cloud sentiment analysis classifies customer phrases only; agent phrases are ignored in the sentiment trend calculation.

    Understand sentiment analysis · article body, description of sentiment trend calculation · Checked 2026-10-01
34. 34
    The Genesys Cloud overall sentiment score ranges from -100 to +100 and gives greater weight to later moments of the interaction.

    Understand sentiment analysis · article body, overall sentiment score description · Checked 2026-10-01
35. 35
    Genesys Cloud speech and text analytics is enabled through a Speech and Text Analytics upgrade add-on to the organization.

    About speech and text analytics · article body, link 'Add Speech and Text Analytics upgrade add-on to your organization' · Checked 2026-10-01
36. 36
    Genesys Cloud Topic Miner analyzes transcripts to discover topics and phrases and can create or update topics from mined data or import topics from JSON files.

    About programs, topics, and phrases · article body, Topic Miner paragraph · Checked 2026-10-01
37. 37
    In a December 20, 2024 release note, Google expanded Quality AI and topic inference beyond English to 38 Gemini languages in preview.

    Customer Experience Insights release notes · release-note entry dated December 20, 2024 · Checked 2026-10-01
38. 38
    Google renamed Conversational Insights to Customer Experience Insights (CX Insights), announced in the release notes on January 12, 2026.

    Customer Experience Insights release notes · release-note entry dated January 12, 2026 · Checked 2026-10-01
39. 39
    Google Customer Experience Insights measures only end-user (customer) sentiment in agent-customer conversations, not agent sentiment.

    Sentiment analysis · page body, scope statement · Checked 2026-10-01
40. 40
    Google Customer Experience Insights sentiment scores lie in the open range (-1, 1); conversation-level sentiment is positive at 0.5 or above, negative at -0.5 or below, and neutral in between.

    Sentiment analysis · sentiment score and conversation-level threshold table · Checked 2026-10-01
41. 41
    Real-time speech analytics generally trades capability for latency: on Amazon Connect, real-time call analysis lacks redaction and generative natural-language category matching that post-call analysis has.inferred

    Automatically categorize contacts by matching conversations with natural language statements, or specific words and phrases · Step 1 Note on Natural Language - Semantic Match; combined with supported-languages footnote ** · Checked 2026-10-01
42. 42
    Sentiment scores are not comparable across CCaaS platforms without normalization, because vendors use different scales (Amazon Connect -5 to +5, Genesys Cloud -100 to +100, Google (-1, 1), NICE categorical) and different speaker scope.inferred

    Investigate sentiment scores during contact conversations using conversational analytics · section 'How sentiment scores are determined' (Amazon scale); other scales per the claims named in notes · Checked 2026-10-01
43. 43
    Microsoft characterizes call-center audio as low quality (8 kHz, possibly two speakers on one channel, possibly compressed) and rates expected recognition quality as medium, under 30% WER.

    Test accuracy of a custom speech model · section 'Example scenario outcomes', both tables, Call center row · Checked 2026-10-01
44. 44
    Microsoft attributes many deletion errors to weak audio signal, insertion errors to noise and crosstalk, and substitution errors to insufficient domain-specific terms.

    Test accuracy of a custom speech model · section 'Resolve errors and improve WER', second paragraph · Checked 2026-10-01
45. 45
    Microsoft defines word error rate (WER) as insertions plus deletions plus substitutions divided by the number of words in the human-labeled reference transcript.

    Test accuracy of a custom speech model · section 'Evaluate word error rate (WER)' · Checked 2026-10-01
46. 46
    Microsoft guidance treats WER of 5-10% as good quality, 20% as acceptable but possibly needing training, and 30% or more as poor quality requiring customization.

    Test accuracy of a custom speech model · section 'Resolve errors and improve WER' · Checked 2026-10-01
47. 47
    NICE CXone Interaction Analytics determines beginning sentiment from the first 400 words or first 30% of the interaction, whichever comes first, and end sentiment from the last 30%.

    Key Terms and Metrics - Interaction Analytics · term headings Beginning Sentiment and End Sentiment · Checked 2026-10-01
48. 48
    NICE CXone Interaction Analytics supports up to 2,000 categories per system.

    Categories - Interaction Analytics · categories overview, system limit statement · Checked 2026-10-01
49. 49
    NICE CXone Interaction Analytics classifies sentiment as positive, negative, mixed or neutral based on counts of positive and negative cues in the transcript.

    Key Terms and Metrics - Interaction Analytics · term headings Positive, Negative, Mixed and Neutral Sentiment · Checked 2026-10-01
50. 50
    In NICE CXone Interaction Analytics, changes to categories applied to workspaces take effect on data only after an administrator clicks Update Data on the Language Model page to queue reprocessing.

    Categories - Interaction Analytics · categories overview, Update Data note · Checked 2026-10-01
51. 51
    In Advanced Sentiment on Webex WFO, only customer sentiment is searchable in Interaction Search, and legacy sentiment is no longer displayed for new interactions once Advanced Sentiment is enabled.

    Advanced Sentiment on Webex WFO (Webex Contact Center) · limitations and enablement notes · Checked 2026-10-01
52. 52
    Advanced Sentiment on Webex WFO for Webex Contact Center analyzes sentiment separately for customers and agents on voice interactions.

    Advanced Sentiment on Webex WFO (Webex Contact Center) · article overview · Checked 2026-10-01
53. 53
    Advanced Sentiment on Webex WFO requires the WFO Bundle SKU, the Analytics + Transcription SKU or the Enterprise Analytics SKU; sentiment trend tracking needs Enterprise Analytics or WFO Bundle plus the 100 Interactions add-on.

    Advanced Sentiment on Webex WFO (Webex Contact Center) · prerequisites section · Checked 2026-10-01
54. 54
    At setup, Webex Contact Center Topic Analytics analyzes the 5,000 most recent interactions or lets the customer upload up to 50,000 historical interactions.

    Get started with Topic Analytics · section on initial topic generation · Checked 2026-10-01
55. 55
    Webex Contact Center Topic Analytics currently supports only English voice transcriptions.

    Get started with Topic Analytics · prerequisites section · Checked 2026-10-01
56. 56
    Webex Contact Center Topic Analytics assigns a topic only to transcripts with at least 25 words and at least 3 conversational turns.

    Get started with Topic Analytics · transcript requirements section · Checked 2026-10-01
57. 57
    Webex Contact Center Topic Analytics requires the AI Assistant add-on SKU A-FLEX-AI-ASST.

    Get started with Topic Analytics · prerequisites section, licensing · Checked 2026-10-01
58. 58
    Cisco recommends at least 2,000 and ideally 7,000 interactions before Webex Contact Center Topic Analytics produces a stable topic list.

    Get started with Topic Analytics · section on initial topic generation and interaction volume · Checked 2026-10-01

Documents

Cite this page

WarmTransfer. (2026, October 1). Speech and interaction analytics. WarmTransfer. https://warmtransfer.net/knowledge/speech-interaction-analytics

@misc{warmtransfer-speech-interaction-analytics,
  title  = {Speech and interaction analytics},
  author = {{WarmTransfer}},
  year   = {2026},
  url    = {https://warmtransfer.net/knowledge/speech-interaction-analytics},
  note   = {Verified 2026-10-01}
}