Speech translation

Translate a finished audio file into text in another language with the OpenAI-compatible translations endpoint.

POST /v1/audio/translations takes a finished audio file and returns its content as text in another language. It follows the OpenAI translations shape. target_language sets the language of the returned text. Models lists the models that translate. The examples use symphony and abbreviate usage.

Translate a file

Upload the audio as multipart/form-data. Without target_language, the text comes back in English.

curl https://api.sprag.ai/v1/audio/translations \
  -H "Authorization: Bearer $SPRAG_API_KEY" \
  -F [email protected] \
  -F model=symphony

{
  "text": "Hello, I would like to reserve a table for four people on Friday evening at around eight o'clock.",
  "usage": { "input": { "audio": { "duration": "6.08" } } }
}

Choose the output language

To get text in a language other than English, send its tag as target_language. To send it from the OpenAI Python client, pass it in extra_body.

curl https://api.sprag.ai/v1/audio/translations \
  -H "Authorization: Bearer $SPRAG_API_KEY" \
  -F [email protected] \
  -F model=symphony \
  -F target_language=de

{
  "text": "Guten Tag, ich hätte gerne einen Tisch für vier Personen am Freitagabend gegen zwanzig Uhr.",
  "usage": { "input": { "audio": { "duration": "6.08" } } }
}

Name the spoken language

If you know the spoken language, send its tag as source_languages.

curl https://api.sprag.ai/v1/audio/translations \
  -H "Authorization: Bearer $SPRAG_API_KEY" \
  -F [email protected] \
  -F model=symphony \
  -F source_languages=fr \
  -F target_language=es

{
  "text": "Hola, me gustaría reservar una mesa para cuatro personas el viernes por la noche, alrededor de las ocho.",
  "usage": { "input": { "audio": { "duration": "6.08" } } }
}

If the request carries one tag, the audio is translated from that language. If it carries several, the spoken language is detected from among them. Send the field once per tag. Without source_languages, the spoken language is detected from the audio.

Request fields

Field	Required	Type	Notes
`file`	Yes	file	The audio to translate. Common containers such as MP3, WAV, FLAC, M4A, OGG, AAC, AIFF, and WebM are accepted.
`model`	Yes	string	A model that translates, such as `symphony`.
`target_language`	No	string	ISO-639 tag of the language to return, such as `de`. English when omitted.
`source_languages`	No	string, repeatable	ISO-639 tag of a language the audio may be spoken in, such as `fr`.
`response_format`	No	string	`json`, the only accepted value.

Target languages

target_language accepts the tags below. Two- and three-letter ISO-639 tags are both accepted, so fr and fra name the same language.

Tag	Language	Tag	Language
`ar`	Arabic	`ms`	Malay
`de`	German	`nl`	Dutch
`en`	English	`pt`	Portuguese
`es`	Spanish	`ru`	Russian
`fr`	French	`tr`	Turkish
`hi`	Hindi	`ur`	Urdu
`id`	Indonesian	`vi`	Vietnamese
`it`	Italian	`yue`	Cantonese
`ja`	Japanese	`zh`	Chinese
`ko`	Korean

If target_language names a language outside this list, the request returns HTTP 400, and the error message lists the accepted tags.

Response fields

Field	Type	Notes
`text`	string	The translated text.
`usage`	object	Measured per-modality usage and cost.

Errors

Status	Meaning
`400`	Sent when `target_language` is not an accepted tag. Also sent when `response_format` is not `json`. Also sent with the code `model_not_supported` when the model does not translate. Also sent with the code `unsupported_media` when the audio cannot be decoded.
`401`	Missing, invalid, or expired credentials.
`402`	No payment method on the account.
`403`	Not a member of the requested organization.
`404`	Unknown model id.
`413`	Request too large.
`429`	Rate limit exceeded. Wait `Retry-After` seconds.
`500`	Unhandled server error.
`503`	Temporarily unavailable. Retry after `Retry-After` when it is set.

Full schemas are in the translations reference. For transcripts in the spoken language, see REST.