Speech-to-Text · Multilingual
Japanese
Language
Benchmarks
XYCER
Character error rate — word scoring mismeasures this language.
AssemblyAIUniversal-3.5 Pro3.4%
ModulateVelma 2 multilingual4.8%
Smallest AIPulse6.1%
OpenAIGPT Live Transcribe6.7%
0.0%20.0%40.0%60.0%
Finalize latency
lower is betterTime from the caller stopping to the finalized transcript, not the first partial — the agent cannot reply until it lands.
AssemblyAIUniversal-3.5 Pro568ms · 1.06s
Smallest AIPulse824ms · 8.25s
0ms300ms600ms900ms1.20s
p50 p90
Time to first token
lower is betterHow long before the first word of the transcript arrives.
AssemblyAIUniversal-3.5 Pro2.36s · 4.46s
Smallest AIPulse3.57s · 4.58s
0ms2.00s4.00s6.00s8.00s
p50 p90
Cost
lower is better$ per 1,000 minutes of audio · vendor list price, the same in every language