AssemblyAI STT 3.7 Realtime Agent vs Fish Audio Transcribe Stream 1

Compare

Accuracy and speed

Left gets fewer words wrong. Lower has the final text ready sooner.

Accurate and fast0ms200ms400ms600ms800ms0%2%4%6%Words wrong on clean audio (Pipecat WER)
  • STT 3.7 Realtime Agent1.68% · 47ms
  • Transcribe Stream 11.90% · 123ms

AssemblyAI STT 3.7 Realtime Agent is ahead on time to first text and final-text delay, and Fish Audio Transcribe Stream 1 on both datasets together. The Pipecat Dataset and The Ocular Dataset are too close to call.

Last updated Methodology by Dileep Chagam

Who is better at what

STT 3.7 Realtime AgentTranscribe Stream 1
  • Overall WER2.41%2.21%
  • Pipecat Dataset WER1.68%1.90%
  • Ocular Dataset WER3.81%2.78%
  • Time to first text493ms730ms
  • Final-text delay47ms123ms

Lower is better on every measure. Too close to call means the likely ranges overlap.

Clip by clip

On every clip both models were scored on, the lower WER wins, and WERs within 5 points of each other are a draw; in the rating, a wider win counts for more. Expected win rate is what the two ELO ratings predict for a match.

STT 3.7 Realtime AgentTranscribe Stream 1
  • ELO rating10241001–1039 · rank 1–1610411035–1046 · rank 1–7
  • Expected win rate48%52%
  • Ocular turns WER · 173 draws of 19810 won15 won
  • Pipecat WER · 903 draws of 99950 won46 won
  • Semantic WER · 193 draws of 1981 won4 won
  • Phone line · 166 draws of 19816 won16 won
  • Call-center noise · 123 draws of 19735 won39 won
  • Babble noise · 113 draws of 19750 won34 won
  • TV noise · 150 draws of 19622 won24 won
  • Reverb · 159 draws of 19816 won23 won
  • Short replies · 63 draws of 703 won4 won
  • Entities · 46 draws of 502 won2 won

Short replies are decided on the error count and entities on exact capture, with no draw margin. How the rating is worked out

STT 3.7 Realtime Agent vs Transcribe Stream 1 FAQ

  • Fish Audio Transcribe Stream 1 is more accurate on the public clips. On the 1,000 public Pipecat clips, AssemblyAI STT 3.7 Realtime Agent scores 2.41% and Fish Audio Transcribe Stream 1 2.21%, so Fish Audio Transcribe Stream 1 makes fewer errors by 0.21 points. On the 300 licensed Ocular turns, AssemblyAI STT 3.7 Realtime Agent scores 1.68% and Fish Audio Transcribe Stream 1 1.90%, with likely ranges of 1.47 to 1.90% and 1.68 to 2.14%. The ranges overlap, so the difference is too close to call. Word error rate counts incorrect, missing, and extra words, so lower is better.

  • Yes. AssemblyAI STT 3.7 Realtime Agent finishes the transcript 237ms sooner. AssemblyAI STT 3.7 Realtime Agent has the final transcript 493ms after the speaker stops and Fish Audio Transcribe Stream 1 730ms, at the median. Both are measured on the same 206 streamed turns.

  • AssemblyAI STT 3.7 Realtime Agent finishes the transcript 237ms sooner, and the two are too close to call on real conversations. A voice agent cannot reply until the final transcript arrives, and the Ocular recordings are real conversations.

  • Yes, both return partial text before the speaker stops. AssemblyAI STT 3.7 Realtime Agent shows first text after 3.81% and Fish Audio Transcribe Stream 1 after 2.78%, at the median.

Test your voice agent the same way.

Cekura runs these simulated calls on your agent and scores every one.