AutoTranscribe & AutoSummary Self-Hosted
ASAPP AutoTranscribe and AutoSummary is a bundled speech-to-text transcription and Summarization, Analytics and Insights service that works with audio recordings/transcriptions of real-time or recorded calls.
Overview
Enterprises need to ensure data privacy and security, comply with regulatory requirements, and maintain control over sensitive information. ASAPP’s SaaS/Cloud or On-premise/Self-Hosted solutions offer enhanced customization, meeting specific organizational needs while seamlessly integrating with existing infrastructure.
AutoTranscribe and AutoSummary enables accurate transcription of voice interactions between contact center agents and customers both in real-time (as calls are occurring, so transcripts are streamed live and other outputs are generated automatically within seconds of the call completion) and in batch (where a set of recorded calls can be operated on and all the outputs are produced in a short period of time). It supports various use cases including fully customizable free text summaries, company-specific call Intent Identification with multiple topics addressed per interaction, rich sentiment analysis and other intelligence about calls that let you cover policy compliance, quality assurance, and coaching use cases.
Additionally, AutoSummary generates targeted structured data to optimize operations and maximize revenue generation for enterprise customers. It efficiently covers 100% of interactions, eliminating the need for agent note-taking or other actions during and after calls.
AutoTranscribe (AT) & AutoSummary (AS) SaaS or self-hosted deployment provides easy integration and the most flexible deployment options in the industry.
Basic Offering
| Product | Features | Real-Time Streaming | Batch Processing | Saas / Cloud Hosting | Self-Hosted (Your Cloud On-Prem) |
|---|---|---|---|---|---|
| AutoTranscribe | Transcription with non-custom (standard) model | ✓ | ✓ | ✓ | ✓ |
| AutoSummary | Summaries with non-custom summaries | ✓ | ✓ | ✓ | ✓ |
Premium Offering
| Product | Features | Real-Time Streaming | Batch Processing | Saas / Cloud Hosting | Self-Hosted (Your Cloud On-Prem) |
|---|---|---|---|---|---|
| AutoTranscribe | Transcriptions with customized (trained on custom vocabulary) models | ✓ | ✓ | ✓ | ✓ |
| AutoSummary |
| ✓ | ✓ | ✓ | ✓ |
Breakdown
ASAPP’s AutoTranscribe service utilizes a speech recognition model, which seamlessly converts spoken words into written text in real-time, complete with appropriate punctuation and capitalization.
AutoSummary uses the capabilities of ASAPP AI services to offer text summarization, Structured Data outputs, and call Intent Classification.
To optimize performance, the model can be customized to support domain-specific needs by training on historical call audio and adding custom vocabulary to further boost recognition accuracy.
The AutoTranscribe & AutoSummary solution pattern consists of the following components to receive speech audio and call signals, and return call transcripts.
- ASAPP works with you to understand your current telephony infrastructure and ecosystem
- Your ASAPP account team will also determine the main use case(s) for the transcript data to determine where and how call transcripts should be sent.
High level architecture for real-time deployment:
High level architecture for batch deployment:
Configuration
AutoTranscribe
Custom vocabulary specification: AutoTranscribe can be taught how to accurately transcribe words that are specific to your business (e.g. product names or technical terminology that is not in common spoken language)
AutoSummary
Structured data specification: The set of structured data produced by Autosummary including entities to be extracted, questions to be answered, etc. can be specified to match the use case (e.g. we have handled compliance, product quality and agent coaching use cases by leveraging structured data in the past).
Free Text Summary Customization: We can customize the free text summaries produced by AutoSummary to include or exclude specific information - for example we could explicitly include promises made by an agent, information on promotions offered, etc.
Input and Output
AutoTranscribe
Input
We support AutoTranscribe via Websocket. Your ASAPP team can work with your organization for other integration options for common telephony technologies such as:
- SIPREC
- Twilio
- Amazon Connect
- Genesys Audiohook
Output
Customers will receive transcripts through the Websocket via Callback, and are responsible for the redaction and persistence of the TTS handlers
AutoSummary
Input
Customers should send a REST API request with full/partial chat transcripts.
Output
Customers receive a response with:
- Summaries
- Standard Free Text Summaries
- Custom Free Text Summaries (only for premium)
- Structured Data
- Including answers to yes/no questions, as well as entities
- Intent (only for premium)
Requirements
ASAPP’s AutoTranscribe service utilizes a speech recognition model, which seamlessly converts spoken words into written text in real-time, complete with appropriate punctuation and capitalization. AutoSummary uses the capabilities of ASAPP AI services to offer text summarization, Structured Data outputs, and call Intent Classification.
To optimize performance, the model can be customized to support domain-specific needs by training on historical call audio and adding custom vocabulary to further boost recognition accuracy.
The following are required for self-hosting AT/AS services.
- Kubernetes cluster with version in maintenance support
- Namespace in Kubernetis cluster
- Kubernetes services to support scheduling and autoscaling:
- nvidia/k8s-device-plugin (Scheduling/Autoscaling)
- nvidia/dcgm-exporter (Autoscaling)
- custom-metrics-apiserver (Autoscaling)
- Prometheus (Metrics)
- Access to NVidia A10G Tensor Core GPUs (e.g. g5.2xlarge on AWS, NVadsA10 on GCP, and v5-series on Microsoft)
- Secure storage for license files (JSON) and rendering into a Kubernetes Secret object
- Networking access for license checks
ASAPP AT/AS services are deployed through OCI Helm charts.
Maintenance
ASAPP Provides:
- Model updates for the speech recognition model, at least twice/year for the first year, and annually thereafter (note this may be limited by the customer’s ability to provide sample outputs)
- Model updates for the free text summarization, structured data and intent model, at least twice / year for the first year, and annually thereafter (note this may be limited by the customer’s ability to provide sample outputs)
- Security patches as needed
- Handling support tickets with standard ASAPP enterprise SLAs (note this may be limited by the customer’s ability to provide logs / telemetry and inputs that reproduce problems)
Customers Provide:
- Sample model outputs each month to assist in debugging and model improvements
- Collections of log data, telemetry on performance and providing this to ASAPP each month to assist in debugging / performance tuning and other system improvements
- Collections of specific inputs associated with unexpected errors observed in the system
Use Cases
Self-hosted deployments are a good fit for customers with high technical complexity. ASAPP is excited to engage with customers to determine the best strategy for our self-hosted offering.
Some use cases include:
- Customers who prefer internal security controls and redaction tooling.
- Customers in highly regulated industries we don’t support on SAAS (e.g. finance or healthcare).
Please don’t hesitate to reach out to ASAPP support!
Was this page helpful?