Features

​

At a glance

​

Streaming

Audio is sent as soon as it is generated, so a caller hears the answer quickly even for long texts. If the client closes the connection, for example when the caller interrupts, synthesis stops.

​

Languages

Turkish, English, German, French, Spanish, Italian and Russian. language sets it; auto detects it from the text. For one-word inputs pass language, since a single word may not show the language.

​

Voices

Eighteen ready voices, by id or name. See Voices.

​

Custom voices

Make a voice from 10-30 s of one person’s clean speech, with their consent. The result is a similar-sounding voice, not an exact copy. See Custom voices.

​

Text normalization

On by default. Numbers, amounts, dates, times and abbreviations are written out before synthesis (“14.30” becomes “on dört otuz”), so send text the way it is written. normalize=false reads the text as it is.

​

Generate and verify

verify=true (Turkish): a speech recogniser listens to each sentence chunk; a chunk it does not hear as written is said once more and the better take is sent. Fewer misread or skipped words, but the first audio comes about 2 s later. Use it for files and prepared announcements, not for live calls.

​

Speed

speed from 0.25 to 4.0 changes the tempo; the pitch stays the same.

​

Audio formats

mp3, opus, aac, flac and speed other than 1 add about 0.07 s to the first audio.

​

Server-Sent Events

stream_format=sse sends speech.audio.delta events with base64 audio, then speech.audio.done.

​

OpenAI compatibility

Any OpenAI TTS client works with only base_url and api_key changed; OpenAI voice names work too. See Text to speech.