Skip to content

Speech models

Speech recognition, text-to-speech, and speech-to-speech quality.

Snapshot: September 20, 2026

Speech-to-text

Artificial Analysis WER Index · Lower is better

Eligible
0
Candidates
71
Coverage
0.0%

unavailable

The Free API rounds AA-WER too coarsely to produce a reliable ranking for this snapshot.

Public site top 3

Read from the Speech-to-text leaderboard on 2026-09-20. This is not part of the dated Free API snapshot.

Rank Model Creator AA-WER · Lower is better
1 Fun-Realtime-ASR-preview Alibaba Cloud 1.7
2 StepAudio 3 ASR StepFun 1.7
3 MAI-Transcribe-2 Microsoft AI 2

Text-to-speech

Text-to-speech Elo · Higher is better

Eligible
90
Candidates
90
Coverage
100.0%
Rank Model Creator Text-to-speech Elo Since prior snapshot
1 Sonic 3.6 Cartesia 1,276 Not comparable
2 Qwen-Audio-3.0-TTS-Plus Alibaba 1,260 Not comparable
3 Realtime TTS-2 Inworld 1,247 Not comparable

Speech-to-speech BBA

BBA score · Higher is better

Eligible
39
Candidates
44
Coverage
88.6%
Rank Model Creator BBA score Since prior snapshot
1 StepAudio 3 Realtime StepFun 1 Not comparable
2 tie Qwen3.5 Omni Plus Realtime Alibaba 0.99 Not comparable
2 tie Qwen Audio 3.0 Realtime Plus Alibaba 0.99 Not comparable

Speech-to-speech FDB

FDB score · Higher is better

Eligible
31
Candidates
44
Coverage
70.5%
Rank Model Creator FDB score Since prior snapshot
1 StepAudio 3 Realtime StepFun 0.99 Not comparable
2 Qwen Audio 3.0 Realtime Plus Alibaba 0.98 Not comparable
3 tie GPT-Live-1 (Sol, low) OpenAI 0.97 Not comparable
3 tie Qwen Audio 3.0 Realtime Flash Alibaba 0.97 Not comparable

Speech-to-speech Tau Voice

Tau Voice score · Higher is better

Eligible
25
Candidates
44
Coverage
56.8%
Rank Model Creator Tau Voice score Since prior snapshot
1 Gemini 3.8 Live Extended Thinking (High) Google 0.69 Not comparable
2 GPT-Live-1 (Astra, medium) OpenAI 0.68 Not comparable
3 GPT-Live-1 (Sol, low) OpenAI 0.59 Not comparable

Data source: Artificial Analysis .