Voice AI BenchmarksBy

ALIBABATEXT-TO-SPEECH5,760 SAMPLES / 30 DAYSLAST RUN OCT 16, 2026, 00:00 UTC

official resource

Qwen3 TTS 1.7b text-to-speech benchmarks

Qwen3 TTS 1.7b, hosted by Baseten (dedicated inference), measures mean 107 ms time to first audio (5th of 32) and 1.8% word error rate (15th of 32) among TTS systems. Results cover the last 30 days. Coval benchmarks hosted endpoints daily using a consistent evaluation methodology. Last measured .

Qwen3-TTS 1.7B is Alibaba's open-weight text-to-speech model in the Qwen3 family, able to clone a voice from a short reference clip.

Time to First Audio#5 / 32
107ms
TTFA Network Roundtrip#4 / 32
74ms
TTFA Leading Silence#10 / 32
33ms
Word Error Rate#15 / 32
1.8%

Overview

The 1.7B-parameter model streams 24 kHz audio over a WebSocket session and clones a voice from roughly ten to twenty seconds of reference audio; Baseten serves it with vLLM on H100 GPUs.

Qwen3 TTS 1.7b 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
Alibaba
Hosted by
Baseten
Source
Dedicated inference
Licensing
Open-weight
Deployment
On-prem
Region
US
Features
Multilingual, Voice cloning

How Qwen3 TTS 1.7b 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 Qwen3 TTS 1.7b highlighted.
  1. #1vui57 ms
  2. #2Default63 ms
  3. #3Qwen3 TTS Fast67 ms
  4. #4TTS Flash 288 ms
  5. #5Qwen3 TTS 1.7b107 ms
  6. #6Simba 3.0110 ms
  7. #7Palabra TTS121 ms
  8. #8Simba 3.2125 ms
  9. #9drama3-preview185 ms
  10. #10TTS 2186 ms
  11. #11Flash v2.5193 ms
  12. #12Eleven v4 Turbo199 ms
Show all 32 models
  1. #13Flux TTS204 ms
  2. #14TTS RT v1255 ms
  3. #15TTS Rt v2256 ms
  4. #16Phantom Z 3.4 conversational268 ms
  5. #17Falcon 2282 ms
  6. #18Mist v3287 ms
  7. #19Sonic 3.5292 ms
  8. #20Aura 2297 ms
  9. #21Sonic 3.6297 ms
  10. #22Coda332 ms
  11. #23S2.1 Pro343 ms
  12. #24Grok TTS361 ms
  13. #25S1391 ms
  14. #26Lightning v3.1 Pro392 ms
  15. #27Daytona TTS485 ms
  16. #28Aura 2 En565 ms
  17. #29Chirp 3 HD598 ms
  18. #30Qwen3 TTS Flash Realtime719 ms
  19. #31S2.1 Pro Free849 ms
  20. #32GPT-4o mini TTS2568 ms
median of all models · 275 ms
TTS models over the last 30 days, ranked on Time to First Audio.
#ModelHostTTFAWERSamples
1vuiFluxions57 ms1,964
2Gradium TTS Beta 202609Gradium63 ms2,136
3Qwen3 TTS FastNari Labs67 ms2,136
4TTS Flash 2Inworld AI88 ms1,940
5Qwen3 TTS 1.7bBaseten107 ms1,440
6Simba 3.0SpeechifyAI110 ms2,133
7Palabra TTS v1Palabra121 ms1,759
8Simba 3.2SpeechifyAI125 ms2,130
9Drama 3 Preview Us EastFish Audio185 ms382
10TTS 2Inworld AI186 ms1,940
11Flash v2.5ElevenLabs193 ms2,126
12Eleven v4 TurboElevenLabs199 ms1,446
13Flux TTSDeepgram204 ms1,991
14TTS RT v1Soniox255 ms1,804
15TTS Rt v2Soniox256 ms1,801
16Phantom Z 3.4 conversationalDeepdub268 ms1,871
17Falcon 2Murf282 ms2,135
18Mist v3Rime287 ms602
19Sonic 3.5Cartesia292 ms2,136
20Aura 2Deepgram297 ms2,068
21Sonic 3.6Cartesia297 ms2,136
22CodaRime332 ms602
23S2.1 ProFish Audio343 ms2,136
24Grok TTSSpaceXAI361 ms2,131
25S1Fish Audio391 ms2,135
26Lightning v3.1 ProSmallest392 ms2,136
27Daytona TTSGuava485 ms1,344
28Aura 2 EnCloudflare565 ms2,135
29Chirp 3 HDGoogle598 ms2,136
30Qwen3 TTS Flash RealtimeAlibaba719 ms2,135
31S2.1 Pro FreeFish Audio849 ms2,129
32GPT-4o mini TTSOpenAI2568 ms2,014
Under-sampled models are excluded; tied models share a place. Dotted WER values split into substitutions, deletions and insertions on hover or tap.

Qwen3 TTS 1.7b runs on dedicated inference, ranked here against shared endpoints.

Latency vs accuracy

TTFA vs WEREach point is one measured model · Qwen3 TTS 1.7b highlighted · 30-day averagesTime to First Audio against Word Error Rate for every measured TTS model, with Qwen3 TTS 1.7b highlighted.
vui — 57 ms, 2.1%Default — 63 ms, 1.9%Qwen3 TTS Fast — 67 ms, 1.8%TTS Flash 2 — 88 ms, 1.7%Simba 3.0 — 110 ms, 1.6%Palabra TTS — 121 ms, 2.7%Simba 3.2 — 125 ms, 2.0%drama3-preview — 185 ms, 1.9%TTS 2 — 186 ms, 1.6%Flash v2.5 — 193 ms, 1.8%Eleven v4 Turbo — 199 ms, 1.7%Flux TTS — 204 ms, 1.7%TTS RT v1 — 255 ms, 2.5%TTS Rt v2 — 256 ms, 2.4%Phantom Z 3.4 conversational — 268 ms, 1.9%Falcon 2 — 282 ms, 1.7%Mist v3 — 287 ms, 2.1%Sonic 3.5 — 292 ms, 1.6%Aura 2 — 297 ms, 1.9%Sonic 3.6 — 297 ms, 1.5%Coda — 332 ms, 9.5%S2.1 Pro — 343 ms, 1.6%Grok TTS — 361 ms, 1.6%S1 — 391 ms, 1.8%Lightning v3.1 Pro — 392 ms, 1.6%Daytona TTS — 485 ms, 1.7%Aura 2 En — 565 ms, 2.1%Chirp 3 HD — 598 ms, 1.6%Qwen3 TTS Flash Realtime — 719 ms, 3.0%S2.1 Pro Free — 849 ms, 1.5%GPT-4o mini TTS — 2568 ms, 1.8%Qwen3 TTS 1.7b — 107 ms, 1.8%

Where the errors come from

WER compositionQwen3 TTS 1.7b's Word Error Rate split by error type · 30-day averageQwen3 TTS 1.7b's WER split into substitutions, deletions and insertions.
  • Qwen3 TTS 1.7b1.8%
SubstitutionsDeletionsInsertions

Averages and tail latency

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

Qwen3 TTS 1.7b latency distributionMilliseconds · shared axis across metrics · last 30 daysQwen3 TTS 1.7b's latency percentiles per metric: p25–p75 band, p50 tick, whisker to max.
  • Time to First Audiop50 104 ms · Max 232 ms
  • TTFA Network Roundtripp50 72 ms · Max 192 ms
  • TTFA Leading Silencep50 30 ms · Max 145 ms
band p25–p75 · tick p50 · whisker to max with p90 and p95 stops
Average and percentile values per metric, with the number of samples behind each row.
MetricAveragep25p50p75p90p95MaxSamples
Time to First Audio107 ms94 ms104 ms117 ms131 ms139 ms232 ms1,440
TTFA Network Roundtrip74 ms68 ms72 ms79 ms84 ms88 ms192 ms1,440
TTFA Leading Silence33 ms23 ms30 ms40 ms55 ms63 ms145 ms1,440
Word Error Rate1.8%0.0%0.0%2.3%7.7%13.8%1277.8%1,440

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 daysQwen3 TTS 1.7b's daily median Time to First Audio over the last 30 days.
Qwen3 TTS 1.7b · Oct 2: 113 msQwen3 TTS 1.7b · Oct 3: 100 msQwen3 TTS 1.7b · Oct 4: 111 msQwen3 TTS 1.7b · Oct 5: 114 msQwen3 TTS 1.7b · Oct 6: 99 msQwen3 TTS 1.7b · Oct 7: 100 msQwen3 TTS 1.7b · Oct 8: 112 msQwen3 TTS 1.7b · Oct 9: 96 msQwen3 TTS 1.7b · Oct 10: 100 msQwen3 TTS 1.7b · Oct 11: 96 msQwen3 TTS 1.7b · Oct 12: 103 msQwen3 TTS 1.7b · Oct 13: 104 msQwen3 TTS 1.7b · Oct 14: 110 msQwen3 TTS 1.7b · Oct 15: 103 msQwen3 TTS 1.7b · Oct 16: 113 ms
Word Error Rate — daily averageDaily average · UTC daysQwen3 TTS 1.7b's daily Word Error Rate over the last 30 days.
Qwen3 TTS 1.7b · Oct 2: 4.9%Qwen3 TTS 1.7b · Oct 3: 2.4%Qwen3 TTS 1.7b · Oct 4: 3.5%Qwen3 TTS 1.7b · Oct 5: 2.0%Qwen3 TTS 1.7b · Oct 6: 1.9%Qwen3 TTS 1.7b · Oct 7: 2.3%Qwen3 TTS 1.7b · Oct 8: 1.2%Qwen3 TTS 1.7b · Oct 9: 2.2%Qwen3 TTS 1.7b · Oct 10: 2.7%Qwen3 TTS 1.7b · Oct 11: 2.2%Qwen3 TTS 1.7b · Oct 12: 3.1%Qwen3 TTS 1.7b · Oct 13: 2.8%Qwen3 TTS 1.7b · Oct 14: 15.7%Qwen3 TTS 1.7b · Oct 15: 3.7%Qwen3 TTS 1.7b · Oct 16: 3.2%

How fast is Qwen3 TTS 1.7b?

On Baseten (dedicated inference), Qwen3 TTS 1.7b measures mean 107 ms time to first audio (5th of 32). Last measured 2026-10-16.

How accurate is Qwen3 TTS 1.7b?

On Baseten (dedicated inference), Qwen3 TTS 1.7b measures 1.8% word error rate (15th of 32). Last measured 2026-10-16.

Who hosts Qwen3 TTS 1.7b?

Qwen3 TTS 1.7b is created by Alibaba and served by Baseten. Coval measures each hosted endpoint separately.

Limits of this comparison

Coval measures the 12Hz base streaming variant on a Baseten dedicated endpoint, so latency reflects Baseten's serving configuration.

  • The endpoint is dedicated inference, so its latency is not ranked against shared APIs here. The benchmark measures audible startup and intelligibility on a fixed voice, not clone similarity or every voice the model can produce.

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