Speech and embeddings

Text to speech

POST /v1/audio/speech returns the whole audio file, synchronously.

Python

speech = client.audio.speech.create(
    model="tts-1",
    voice="nova",
    input="Your order is packed and waiting at the pickup point.",
)
speech.write_to_file("out.mp3")

TypeScript

const speech = await client.audio.speech.create({
  model: 'tts-1',
  voice: 'nova',
  input: 'Your order is packed and waiting at the pickup point.',
})
await Bun.write('out.mp3', speech)

cURL

curl https://api.mixen.ai/v1/audio/speech \
  -H "Authorization: Bearer $MIXEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "tts-1", "voice": "nova", "input": "Your order is packed and waiting at the pickup point."}' \
  -o out.mp3

Field	Description
input	The text, required
model	tts-1 is accepted for compatibility; you can also name a catalog model
voice	alloy, echo, fable, nova, shimmer and others — the full list is the voices field in the catalog. An unknown voice silently falls back to the default
response_format	Audio format
speed	Speech rate

Billing is by the length of the input text, per 1000 characters. The duration of the resulting audio does not affect the price, so you can budget before you call.

Transcription

POST /v1/audio/transcriptions takes multipart/form-data.

curl https://api.mixen.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $MIXEN_API_KEY" \
  -F file=@meeting.m4a \
  -F model=whisper-1 \
  -F language=en

Field	Description
file	Audio file, required
model	whisper-1 for compatibility
language	Language code; detected automatically when omitted
response_format	json by default

Billed per minute of audio.

Embeddings

POST /v1/embeddings produces vectors for search, clustering and RAG.

vec = client.embeddings.create(
    model="text-embedding-3-small",
    input=["first document", "second document"],
)
print(len(vec.data), len(vec.data[0].embedding))

Accepted models are text-embedding-3-small, text-embedding-3-large and text-embedding-ada-002. The input field takes a string or an array of strings — an array is cheaper and faster than looping one at a time.

Billed per million input tokens.

Scopes

Speech and transcription need the audio scope; embeddings need embeddings. A key without the right scope gets a 403.