Voice AI BenchmarksBy

FISH AUDIOTEXT-TO-SPEECH40,888 SAMPLES / 30 DAYSLAST RUN AUG 28, 2026, 21:30 UTC

official resource

S2.1 Pro text-to-speech benchmarks

S2.1 Pro is Fish Audio's production endpoint in the S2.1 synthesis family.

Time to First Audio#14 / 30
374ms
TTFA Network Roundtrip#21 / 28
356ms
TTFA Leading Silence#3 / 28
17ms
Word Error Rate#7 / 30
4.6%

Overview

Both S2.1 routes support multilingual synthesis, voice cloning and emotion control and receive identical benchmark prompts.

S2.1 Pro is tested every day on a fixed set of text prompts, measuring how quickly audible speech starts and how intelligible the result is.

Technical specifications

Made by
Fish Audio
Hosted by
Fish Audio
Source
Official API
Licensing
Proprietary
Deployment
Cloud
Region
US
Features
Multilingual, Voice cloning, Emotion control

How S2.1 Pro ranks

Full TTS dashboard
Time to First Audio — every measured TTS modelMilliseconds · lower is better · 30-day averageEvery TTS model ranked on Time to First Audio, with S2.1 Pro highlighted.
  1. #1Palabra TTS v1116 ms
  2. #2vui124 ms
  3. #3TTS Flash 2128 ms
  4. #4TTS 2176 ms
  5. #5Blizzard235 ms
  6. #6Neural236 ms
  7. #7Mist v3256 ms
  8. #8TTS Rt v2262 ms
  9. #9Sonic 3.5274 ms
  10. #10TTS RT v1275 ms
  11. #11Dragon HD Latest310 ms
  12. #12Coda313 ms
  13. #13Aura 2328 ms
  14. #14S2.1 Pro374 ms
Show all 30 models
  1. #15Default384 ms
  2. #16Speech 2.8 Turbo411 ms
  3. #17Eleven v3 Conversational412 ms
  4. #18Grok TTS420 ms
  5. #19S1434 ms
  6. #20Flash v2.5455 ms
  7. #21Speech 2.8 HD460 ms
  8. #22Sonic 3.6466 ms
  9. #23Simba 3.2484 ms
  10. #24Chirp 3 HD512 ms
  11. #25Simba 3.0526 ms
  12. #26Falcon 2549 ms
  13. #27Lightning v3.1 Pro586 ms
  14. #28Qwen3 TTS Flash Realtime645 ms
  15. #29S2.1 Pro Free806 ms
  16. #30GPT-4o mini TTS1075 ms
median of all models · 397 ms
TTS models over the last 30 days, ranked on Time to First Audio.
#ModelHostTTFAWERSamples
1Palabra TTS v1Palabra116 ms5.9%14,396
2vuiFluxions124 ms8,624
3TTS Flash 2Inworld AI128 ms8,719
4TTS 2Inworld AI176 ms4.8%14,529
5BlizzardLmnt235 ms7.4%14,117
6NeuralAzure236 ms4.3%3,350
7Mist v3Rime256 ms6.3%14,524
8TTS Rt v2Soniox262 ms8,849
9Sonic 3.5Cartesia274 ms6.1%14,513
10TTS RT v1Soniox275 ms3.7%14,528
11Dragon HD LatestAzure310 ms5.1%3,350
12CodaRime313 ms5.2%14,523
13Aura 2Deepgram328 ms5.3%14,501
14S2.1 ProFish Audio374 ms10,222
15DefaultGradium384 ms4.6%14,002
16Speech 2.8 TurboMiniMax411 ms4.8%1,997
17Eleven v3 ConversationalElevenLabs412 ms6,750
18Grok TTSxAI420 ms4.7%14,524
19S1Fish Audio434 ms4.9%10,236
20Flash v2.5ElevenLabs455 ms6.7%9,259
21Speech 2.8 HDMiniMax460 ms1,995
22Sonic 3.6Cartesia466 ms560
23Simba 3.2Speechify484 ms4.7%14,525
24Chirp 3 HDGoogle512 ms5.2%14,530
25Simba 3.0Speechify526 ms5.3%14,526
26Falcon 2Murf549 ms10,229
27Lightning v3.1 ProSmallest586 ms4.4%14,515
28Qwen3 TTS Flash RealtimeAlibaba645 ms8.8%14,530
29S2.1 Pro FreeFish Audio806 ms4.7%14,464
30GPT-4o mini TTSOpenAI1075 ms4.8%14,517
Under-sampled models are excluded; tied models share a place. Dotted WER values split into substitutions, deletions and insertions on hover or tap.

Highest relative placement: 7th of 30 on Word Error Rate.

Latency vs accuracy

TTFA vs WEREach point is one measured model · S2.1 Pro highlighted · 30-day averagesTime to First Audio against Word Error Rate for every measured TTS model, with S2.1 Pro highlighted.
Palabra TTS v1 — 116 ms, 5.9%vui — 124 ms, 7.9%TTS Flash 2 — 128 ms, 4.9%TTS 2 — 176 ms, 4.8%Blizzard — 235 ms, 7.4%Neural — 236 ms, 4.3%Mist v3 — 256 ms, 6.3%TTS Rt v2 — 262 ms, 3.9%Sonic 3.5 — 274 ms, 6.1%TTS RT v1 — 275 ms, 3.7%Dragon HD Latest — 310 ms, 5.1%Coda — 313 ms, 5.2%Aura 2 — 328 ms, 5.3%Default — 384 ms, 4.6%Speech 2.8 Turbo — 411 ms, 4.8%Eleven v3 Conversational — 412 ms, 4.4%Grok TTS — 420 ms, 4.7%S1 — 434 ms, 4.9%Flash v2.5 — 455 ms, 6.7%Speech 2.8 HD — 460 ms, 4.7%Sonic 3.6 — 466 ms, 5.5%Simba 3.2 — 484 ms, 4.7%Chirp 3 HD — 512 ms, 5.2%Simba 3.0 — 526 ms, 5.3%Falcon 2 — 549 ms, 5.2%Lightning v3.1 Pro — 586 ms, 4.4%Qwen3 TTS Flash Realtime — 645 ms, 8.8%S2.1 Pro Free — 806 ms, 4.7%GPT-4o mini TTS — 1075 ms, 4.8%S2.1 Pro — 374 ms, 4.6%

Where the errors come from

WER compositionS2.1 Pro's Word Error Rate split by error type · 30-day averageS2.1 Pro's WER split into substitutions, deletions and insertions.
  • S2.1 Pro4.6%
SubstitutionsDeletionsInsertions

Averages and tail latency

Averages hide slow outliers — these are the distributions behind each figure.

S2.1 Pro latency distributionMilliseconds · shared axis across metrics · last 30 daysS2.1 Pro's latency percentiles per metric: p25–p75 band, p50 tick, whisker to p99.
  • Time to First Audiop50 334 ms · p99 964 ms
  • TTFA Network Roundtripp50 318 ms · p99 943 ms
  • TTFA Leading Silencep50 7 ms · p99 108 ms
band p25–p75 · tick p50 · whisker to p99 with p90 and p95 stops
Average and percentile values per metric, with the number of samples behind each row.
MetricAveragep25p50p75p90p95p99Samples
Time to First Audio374 ms300 ms334 ms380 ms474 ms706 ms964 ms10,222
TTFA Network Roundtrip356 ms288 ms318 ms357 ms436 ms691 ms943 ms10,222
TTFA Leading Silence17 ms2 ms7 ms22 ms49 ms76 ms108 ms10,222
Word Error Rate4.6%0.0%0.0%7.7%18.8%21.4%30.0%10,222

Last 30 days

Daily medians from the same measurement runs · gaps are days without qualifying runs.

Time to First Audio — daily p50Line p50 · band p25–p75 · UTC daysS2.1 Pro's daily median Time to First Audio over the last 30 days.
S2.1 Pro · Aug 7: 330 msS2.1 Pro · Aug 8: 323 msS2.1 Pro · Aug 9: 322 msS2.1 Pro · Aug 10: 348 msS2.1 Pro · Aug 11: 374 msS2.1 Pro · Aug 12: 365 msS2.1 Pro · Aug 13: 396 msS2.1 Pro · Aug 14: 398 msS2.1 Pro · Aug 15: 369 msS2.1 Pro · Aug 16: 373 msS2.1 Pro · Aug 17: 364 msS2.1 Pro · Aug 18: 347 msS2.1 Pro · Aug 19: 312 msS2.1 Pro · Aug 20: 296 msS2.1 Pro · Aug 21: 333 msS2.1 Pro · Aug 22: 309 msS2.1 Pro · Aug 23: 315 msS2.1 Pro · Aug 24: 317 msS2.1 Pro · Aug 25: 293 msS2.1 Pro · Aug 26: 292 msS2.1 Pro · Aug 27: 284 msS2.1 Pro · Aug 28: 293 ms
Word Error Rate — daily averageDaily average · UTC daysS2.1 Pro's daily Word Error Rate over the last 30 days.
S2.1 Pro · Aug 7: 4.2%S2.1 Pro · Aug 8: 4.8%S2.1 Pro · Aug 9: 5.6%S2.1 Pro · Aug 10: 4.5%S2.1 Pro · Aug 11: 5.3%S2.1 Pro · Aug 12: 5.1%S2.1 Pro · Aug 13: 4.0%S2.1 Pro · Aug 14: 5.0%S2.1 Pro · Aug 15: 4.4%S2.1 Pro · Aug 16: 5.0%S2.1 Pro · Aug 17: 4.4%S2.1 Pro · Aug 18: 4.7%S2.1 Pro · Aug 19: 4.6%S2.1 Pro · Aug 20: 4.9%S2.1 Pro · Aug 21: 5.1%S2.1 Pro · Aug 22: 4.3%S2.1 Pro · Aug 23: 5.3%S2.1 Pro · Aug 24: 5.0%S2.1 Pro · Aug 25: 5.1%S2.1 Pro · Aug 26: 5.4%S2.1 Pro · Aug 27: 4.4%S2.1 Pro · Aug 28: 4.8%

Time to First Audio by dataset

S2.1 Pro by test conditionMilliseconds · lower is better · best condition firstS2.1 Pro's Time to First Audio on each benchmark dataset.
  1. Text prompts374 ms
Time to First Audio per dataset, with the number of samples behind each figure.
DatasetTTFASamples
Text prompts374 ms10,222

Strongest condition: Text prompts at 374 ms · weakest: Text prompts at 374 ms.

Limits of this comparison

Coval reports it separately from S2.1 Pro Free so service-tier behavior is not averaged together.

  • Tier labels do not guarantee a performance ordering; the current measurements determine relative rank.

Official sources

Results are re-measured daily using fixed datasets and reported over a rolling 30-day window. Same datasets, prompts and metric definitions for every model, measured by Coval’s open-source runner. Full methodology on the overview.

Evaluate your own voice agent

Use Coval to test your production configuration, prompts and calls—not only the public benchmark endpoints.

Book a Demo