customers
All customer stories
Top Voice AI companies are building with Assembly.
resources
Latest Release
Universal 3.5 Pro
Our new flagship speech-to-text model handles real-world audio the way it actually happens. Available for real-time and pre-recorded audio.
resources
Your WER benchmark might be lying to you. Models that look great in demos fail on real audio—noise, numbers, domain terms, overlap, and edge cases. Measure what actually breaks your pipeline.
Start the patient on lisinopril, ten milligrams daily.
Got it—and the follow-up is in two weeks?
Yes. Watch the alphanumeric dosage codes—that's where models slip.
Go beyond WER with domain-specific evals like Missed Entity Rate and Semantic WER, so the score reflects your use case.
WER scores "gonna → going to" the same as "lisinopril → listening a pill." One keeps the meaning; one breaks your pipeline.
Truth File Corrector, semantic word lists, and a benchmarking SDK clean up ground truth and compare models on your own audio.
36%
improvement in close rate
“The new Universal-3.5 Pro speech model from AssemblyAI is best so far in terms of accuracy, latency, and language switching.”
80%
increase in customer satisfaction
“Assembly has saved us countless hours managing models, and provided exceptional accuracy.”
36%
improvement in close rate
“The new Universal-3.5 Pro speech model from AssemblyAI is best so far in terms of accuracy, latency, and language switching.”
80%
increase in customer satisfaction
“Assembly has saved us countless hours managing models, and provided exceptional accuracy.”
Create a free account, drop your audio into the playground, and compare models side by side. Use the benchmarking SDK to run WER, Missed Entity Rate, and Semantic WER against your own ground truth.
Start building freeNo credit card required
Word error rate is the standard measure of transcription accuracy, but it only tells the truth when you run it on real, varied audio—not a clean demo clip.
Pre-recorded word error rate, public benchmark datasets.
*Lower is better*Source: AssemblyAI published benchmarks — assemblyai.com/benchmarks.
