eroq api · voice
Give it a voice.
Then get it back.
Two speech models at 2 to 3 credits per hundred characters, a curated roster, and cloning from a clean recording. Assign a voice to a character and every line they speak is the same voice — then transcribe audio back to text on the same key.
MP3 out · 13 languages · clones private unless you publish them
40 free credits on signup · no card required
curl https://eroq.ai/v1/audio/speech \
-H "Authorization: Bearer $EROQ_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "eroq-voice-one",
"input": "That noise is the reactor missing its coolant flush. [sighing] Again.",
"voice": "aria"
}' \
--output speech.mp3The models
3 models. Flat prices.
eroq Voice One
speecheroq-voice-one
Natural speech synthesis with a curated voice roster. Returns MP3 audio directly.
3 credits / 100 characters
eroq Voice Turbo
speecheroq-voice-turbo
The high-volume voice tier: lower latency and a lower price, same roster and cloned voices as eroq Voice One.
2 credits / 100 characters
Scribe One
transcriptioneroq-scribe-one
Transcription with automatic language detection, built for voice-message-length audio.
1 credit / minute
What you get
Voice, built for products that ship.
Two speeds
eroq Voice One at three credits per hundred characters, eroq Voice Turbo at two. Same request shape and the same roster — Turbo trades a little nuance for latency and cost when the line is a notification rather than a performance.
Clone a voice
Upload a clean recording and the clone is yours alone — private to your account, usable from the API, the studio and any character you assign it to. No one else can reach it unless you publish it to the voice catalog yourself, with a consent attestation and a review first.
Characters that speak
Bind a voice to a saved character and every line they say comes out in it, whether the call comes from your product, a film or the studio. One assignment rather than a voice id threaded through every call site.
Direction in the text
Speed and expressiveness are parameters, and the rest of the performance lives in the input: punctuation — commas, ellipses, line breaks — on both engines, and on eroq Voice One tags in square brackets like [whispering] or [laughing], performed and never read out. There is no SSML dialect to learn and none to get wrong.
Thirteen languages
The same voices speak thirteen languages, so a cloned narrator can read copy that was never recorded in that language. Pronunciation follows the text, so numbers and names are worth spelling out.
And back again
Transcription runs on the same key at one credit per minute of audio, which closes the loop for captioning your own narration. There is no speaker diarization and no word-level timestamps, so it returns text, not a subtitle file.
In the reference
Keep reading
FAQ
The short version.
What does speech cost? +
Three credits per hundred characters on eroq Voice One and two on eroq Voice Turbo, billed on the text you send. Transcription is one credit per started minute of audio.
Who can use a voice I clone? +
Only your account, unless you publish the clone to the public voice catalog (POST /v1/voices/{id}/publish, with a consent attestation and a review). Clones can be unpublished and deleted, which removes them from every character they were assigned to.
What format comes back? +
MP3. There is no streaming audio endpoint, so a long passage arrives as one finished file rather than as it is spoken.
Can it lip-sync to a generated face? +
No. Speech and video are separate calls and there is no lip-sync tool. Veo 3.1 can render a spoken line as part of a clip, but you cannot lay a voice track over an existing clip and have the mouth match.
Does transcription return timestamps? +
No — it returns text. There is no diarization, no word-level timing and no subtitle file, so captions still need a pass in your editor.
First call in five minutes.
Sign up, create a key, paste the quickstart. 40 free credits to start, and plans from $15.