Deepgram has published an evaluation framework for voice AI platforms in contact centres, drawing on analysis of more than 4 million voice interactions to define minimum performance requirements for deployments at scale.
The framework sets language-specific word error rate (WER) targets, a measure of transcription accuracy, requiring English below 10%, German below 12% and Hindi below 15%.
On latency, the standard requires responses within 500 milliseconds for natural conversation flow, with 300 milliseconds preferred, and sets an architectural expectation of 10,000 or more simultaneous calls without performance degradation.
For contact centres handling more than 50,000 calls per month, Deepgram estimates base per-minute costs translate to between $7,500 and $37,500 monthly before accounting for integration, training and compliance expenses.
The evaluation methodology used phased concurrency testing, scaling from 100 to more than 10,000 simultaneous calls, reviewed contractual service level agreements (SLAs) ranging from 99.9% to 99.999% uptime, and assessed platforms against SOC 2 Type II, PCI-DSS and HIPAA compliance requirements, including AES-256 encryption at rest and TLS 1.2 or higher in transit.
The framework also requires vendors to provide load testing results under a non-disclosure agreement as part of the procurement process.
Deepgram ranked itself as the leading platform for production-scale infrastructure in the analysis, citing its Voice Agent API's support for HIPAA and GDPR compliance, cloud and on-premises deployment options, bundled pricing and infrastructure it says handles more than 140,000 concurrent calls.
Related reading
- Deepgram triples concurrency limits for AI voice agents
- New guide ranks leading speech-to-text APIs for voice application developers
- Google for Education to offer free AI training to all 6 million US teachers and university faculty
The release also assessed platform strengths for nine other providers, including Five9, Genesys Cloud CX, NICE CXone, Talkdesk, Amazon Connect and Google Cloud CCAI.
Deepgram is offering teams $200 in free credits to test the Voice Agent API through its Console product before committing to production deployments.
The recap
- Evaluation sets production baselines for voice AI platforms' enterprise deployment.
- 4 million+ calls analyzed; WER targets include English below 10%.
- Test Voice Agent API in Deepgram Console with $200 credits.
