Streams as it renders
Your text is split on sentence boundaries and synthesised chunk by chunk. Playback begins on the first chunk instead of waiting for the whole passage.
Free during beta · credits granted on request
Text to speech that starts talking instantly
Fifty natural voices across nine languages, rendered on our own hardware and streamed sentence by sentence — so the first words are already playing while the rest is still being generated.
No card, no API key. Tell us what you're building and we'll top up your account.
Heart · English (US)
Grade A · 24 kHz mono
chunk 3 / 7streaming
first sound 640 ms · 412 credits · 27s of audio
af_heartbm_georgeff_siwishf_alphajf_alpha
- voices
- 51
- languages
- 9
- output
- 24 kHz
- to first sound
- < 1s
How it works
Four steps from signup to sound
01
Ask for credits
Tell us who you are and what you want to build. We read every request by hand and top up your account — usually the same day.
02
Sign in with your email
No password to remember. We send a one-time link; clicking it drops you straight into the Studio.
03
Paste, pick a voice, generate
Choose a language and voice, tune the speed, and hit generate. Audio starts playing on the first sentence and downloads as a WAV when it's done.
04
Only pay for what renders
One credit is one character. Credits are held up front and settled against the audio actually produced — stop halfway and the rest comes straight back.
What you get
Small, fast and honest about what it costs
Nine languages
English (US and UK), Spanish, French, Hindi, Italian, Portuguese, Japanese and Mandarin — each with voices tuned for it.
Credits you can read
One credit, one character. No opaque token maths — the counter in the editor is the exact price of what you typed.
Speed control
Slow a narration down to 0.5x or push an alert to 2x without changing pitch or re-recording.
Clean WAV out
24 kHz mono PCM, no watermark, no attribution requirement. Download it or wire it straight into your pipeline.
The roster
51 voices, nine languages
Grades come from the Kokoro model card and reflect how much audio each voice was trained on. A and B hold up best across long passages.
Heart
af_heart
Female · Flagship quality
Bella
af_bella
Female · Flagship quality
Nicole
af_nicole
Female · Very good
Aoede
af_aoede
Female · Very good
Kore
af_kore
Female · Very good
Sarah
af_sarah
Female · Very good
Nova
af_nova
Female · Good
Sky
af_sky
Female · Good
Alloy
af_alloy
Female · Good
Jessica
af_jessica
Female · Good
River
af_river
Female · Good
Michael
am_michael
Male · Very good
Fenrir
am_fenrir
Male · Very good
Puck
am_puck
Male · Very good
Echo
am_echo
Male · Good
Eric
am_eric
Male · Good
Liam
am_liam
Male · Good
Onyx
am_onyx
Male · Good
Adam
am_adam
Male · Experimental
Santa
am_santa
Male · Experimental
Voice previews land in the Studio once you're signed in — generating a sample costs the same credits as any other clip.
Pricing
Free while we're in beta
There is no billing in Voicer today. Credits are granted by hand so we can keep the hardware honest and talk to everyone using it. When we do start charging, you'll hear it from us first — and nothing you've already generated will be clawed back.
Open betano card required
Freewhile in beta
- Credits granted on request, topped up on request
- All 50 voices and 9 languages, no tiering
- Speed control from 0.5x to 2x
- WAV download with no watermark
- Direct line to the person who built it
One credit, one character
Credits are held when you press generate and settled against the audio that actually rendered. Cancel mid-stream and the balance comes back.
- a 15-second clip
- ≈ 225
- a minute of narration
- ≈ 900
- about 16 minutes
- ≈ 15,000
We suggest keeping clips around 15 seconds (~225 characters). Longer is allowed — it just costs proportionally more and takes longer to render.
For developers
An HTTP API
Coming soonBot and server access is not open yet — during beta every generation goes through a signed-in session so we can keep the hardware from being scraped. The endpoint shape below is what we're building towards, and the reference is already written.
preview only
curl https://api.strophic.in/v1/speech \
-H "Authorization: Bearer $VOICER_KEY" \
-H "Content-Type: application/json" \
-d '{
"text": "The build finished in ninety seconds.",
"voice": "af_heart",
"speed": 1.0,
"format": "wav"
}' --output alert.wavQuestions
Before you ask
Why do I have to ask for credits?
Voicer runs on a single small server we pay for ourselves. Granting credits by hand keeps that box responsive and lets us actually talk to the people using it. It usually takes less than a day.
How long can a single generation be?
There is no hard cap on voice length beyond a 5,000 character safety ceiling per request. We suggest keeping clips near 15 seconds — longer text works fine, it just costs proportionally more credits and takes longer to render.
What happens when I run out?
The Studio switches to a short form asking what you were making and what would make Voicer worth paying for. Send it and we'll usually top you up.
Can I use the audio commercially?
Yes. Kokoro-82M is Apache 2.0 and we add no restrictions of our own. The audio has no watermark and needs no attribution.
Is there an API?
Not yet. During beta everything goes through a signed-in session. The reference is published so you can see the shape it will take, and we'll open keys once the hardware can take it.
Do you keep my text?
We store the first 160 characters of each generation as a preview so you can find it in your history, plus the character count for billing. The audio itself is never written to disk on our side.
Anything else — write to hello@strophic.in.
Tell us what you're building
Every account is opened by hand. Send a couple of sentences about your project and we'll load your credits — free, for as long as the beta runs.