Text to Speech

Synthesizes text into speech and streams audio chunks as they are generated. The response body is binary audio, not a JSON envelope. Signed, silent provenance metadata is embedded in WAV, MP3, Ogg Opus, WebM Opus, and FLAC output. Raw PCM, mu-law, A-law, and HLS output do not carry provenance metadata.

POST /v1/speech

Text to Speech

cURL

curl --request POST \
  --url https://api.spitch.app/v1/speech \
  --header 'Authorization: Bearer <token>' \
  --header 'Content-Type: application/json' \
  --data '
{
  "text": "<string>",
  "voice": "<string>"
}
'

200

"<string>"

Authorizations

Authorization · string · header · required

Authenticate with Authorization: Bearer <token>. The service accepts JWTs, API keys, and guest tokens through this bearer token header.

Body

application/json

JSON synthesis request. The response is streamed binary audio, not JSON.

text · string · required

Text to synthesize into speech.

voice · string · required

Voice ID to use for synthesis. Use a built-in system voice, a registered custom voice, or blank.

language · string

Optional ISO 639 language code. Yoruba input is diacritized before synthesis.

format · enum<string> · default: mp3

Output audio format. Silent signed provenance metadata is available for WAV, MP3, Ogg Opus, WebM Opus, and FLAC, but not raw PCM, mu-law, A-law, or HLS.

Available options: wav, mp3, ogg_opus, webm_opus, flac, pcm_s16le, mulaw, alaw, m3u8

bitrate · enum<string> · default: 128k

Output bitrate for compressed formats.

Available options: 32k, 48k, 64k, 96k, 128k, 192k

sample_rate · enum<integer> · default: 24000

Output sample rate in hertz. Only 24000 is accepted.

Available options: 24000

speed · number · default: 1

Speech speed multiplier.

Required range: 0.7 <= x <= 1.2

Response

200

audio/wav

Chunked streaming audio. The media type depends on the requested format; formats without a mapped media type are returned as application/octet-stream.

The response is of type file.