Voices

Find a voice

Every voice model Speko can route to, with the measured naturalness behind it.

Browse the roster

The catalog endpoint did not answer, so this is the bundled snapshot — which carries the voice models but not their per-voice rosters. The boards below are unaffected.

No per-voice roster in this snapshot. The models are listed below.

What each one scored

Naturalness is measured per language, not inferred from English — the model that wins one study routinely loses another. Every figure is the board's own.

#ModelNaturalnessarena EloSynthenglish board p50Cost$ / M chars
1
gemini-3.1-flash-tts-previewgoogle-tts:gemini-3.1-flash-tts-preview
1591978ms~$33.3
2
eleven_v3elevenlabs:eleven_v3
1590481ms$100.0
3
aura-2deepgram:aura-2
1584125ms$30.0
4
sonic-3.5cartesia:sonic-3.5
1574121ms$50.0
5
simba-3.2speechify:simba-3.2
1573345ms$10.0
6
tts-rt-v1soniox:tts-rt-v1
1569362ms~$13.0
7
s2.1-probenchmark only — not routable
~1566185ms$15.0
8
inworld-tts-2inworld:inworld-tts-2
1561116ms$25.0
9
lightning_v3.1smallest:lightning_v3.1
1544173ms$25.0
10
palabra-tts-v1benchmark only — not routable
~152072ms$30.0
11
grok-ttsxai:tts
1495272ms$15.0
12
defaultgradium:default
1448244ms$57.8
13
speech-2.8-hdminimax:speech-2.8-hd
1431294ms$100.0
14
arcanav3rime:arcanav3
1429238ms$40.0
15
gpt-4o-mini-ttsopenai:gpt-4o-mini-tts
1424691ms~$20.0
16
octave-2hume:octave-2
1377448ms$100.0
17
qwen3-tts-flashalibaba:qwen3-tts-flash
1310472ms$10.0

Naturalness / arena Elo. Higher is better. Blind A/B votes, field mean 1500. Latency and price are the English board’s, by model — a model measured only in this study carries neither.

Voice models and language coverage

Which model each voice belongs to, and the languages the router will select it for.

ModelVoices
sonic-3.5cartesia:sonic-3.5
—
enardeesfilfrhinbtate
eleven_v3elevenlabs:eleven_v3
—
enardeesfilfrhinbta
speech-2.8-hdminimax:speech-2.8-hd
—
endeesfilfrhinb
gemini-3.1-flash-tts-previewgoogle-tts:gemini-3.1-flash-tts-preview
—
enfilhinbtate
gpt-4o-mini-ttsopenai:gpt-4o-mini-tts
—
endeesfilfrnb
grok-ttsxai:tts
—
enardeesfilfr
inworld-tts-2inworld:inworld-tts-2
—
enardeesfilfr
arcanav3rime:arcanav3
—
endeesfr
aura-2deepgram:aura-2
—
endeesfr
lightning_v3.1smallest:lightning_v3.1
—
endeesfr
octave-2hume:octave-2
—
endeesfr
qwen3-tts-flashalibaba:qwen3-tts-flash
—
endeesfr
tts-rt-v1soniox:tts-rt-v1
—
endeesfr
defaultgradium:default
—
endefr
simba-3.2speechify:simba-3.2
—
en
palabra-tts-v1benchmark only — not routable
—
no languages declared
s2.1-probenchmark only — not routable
—
no languages declared

Take one to the API

A voice is two strings: the model id you send as model, and the voice’s own id you send as voice. Both are in the table above, and the speech endpoint is the OpenAI-compatible one your client already calls.

curl https://api.speko.ai/v1/audio/speech \
  -H "Authorization: Bearer $SPEKO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "auto",
    "voice": "alloy",
    "input": "Hello from Speko."
  }'

Leave model as auto and the router picks the voice model for you, per language and per objective, from the same measurements the boards on this page publish.