AI Research ·
Speech-to-Text on Phone Audio: 10 Models Tested on Names, Numbers and Latency (Nov 2026)
Stork tested 10 speech-to-text models on 16.6 minutes of real caller and accented speech over a simulated G.711 phone line: word error rate, names, digits and addresses, reply time and price. Plus text-to-speech time to first audio for 9 voice models. Run November 11, 2026.