Audio responses

Use this guide for spoken output from a Response. To synthesize supplied text directly, see Text to speech. The OpenAI examples use its custom-request method for MKA1 extensions that are not part of the OpenAI Responses types.

Generate audio (text-to-speech)

Request audio output by setting modalities to ["text", "audio"]. The response includes the text reply and a base64-encoded recording of it.

Audio configuration

Audio output is always a WAV file: 24 kHz, 16-bit mono. The voice follows the language of the reply, which the API detects from the text (see Supported languages). The request’s audio object (voice and format) is accepted but has no effect, so the examples omit it.

import { SDK } from '@meetkai/mka1';
import { writeFileSync } from 'fs';
import type { ResponseObject } from '@meetkai/mka1/models/components';

const sdk = new SDK({ bearerAuth: 'Bearer <mka1-api-key>' });

const result = await sdk.llm.responses.create({
  xOnBehalfOf: '<end-user-id>', // optional — attribute the request to one of your end users
  responsesCreateRequest: {
    model: 'auto',
    input: 'Say hello in a friendly way. Keep it very short.',
    modalities: ['text', 'audio'],
  },
});

// The output includes an output_audio item with base64 data and a transcript

import OpenAI from 'openai';
import { writeFileSync } from 'fs';

const openai = new OpenAI({
  apiKey: '<mka1-api-key>',
  baseURL: 'https://apigw.mka1.com/api/v1/llm/',
  defaultHeaders: { 'X-On-Behalf-Of': '<end-user-id>' },
});

const response = await openai.post<{ output: Array<{ type: string; data?: string; transcript?: string }> }>('/responses', { body: {
  model: 'auto',
  input: 'Say hello in a friendly way. Keep it very short.',
  modalities: ['text', 'audio'],
  stream: false,
} });

// Find the audio output
const audioItem = response.output.find((item) => item.type === 'output_audio');
// audioItem.data contains base64-encoded WAV
// audioItem.transcript contains the spoken text

from meetkai_mka1 import SDK
import base64

sdk = SDK(bearer_auth="Bearer <mka1-api-key>")

result = sdk.llm.responses.create(
    model="auto",
    input="Say hello in a friendly way. Keep it very short.",
    modalities=["text", "audio"],
)

# The output includes an output_audio item with base64 data and a transcript

using MeetKai.MKA1;
using MeetKai.MKA1.Types.Components;
using MeetKai.MKA1.Types.Requests;

var sdk = new SDK(bearerAuth: "Bearer <mka1-api-key>");

var result = await sdk.Llm.Responses.CreateAsync(new ResponsesCreateRequest()
{
    Model = "auto",
    Input = ResponsesCreateRequestInput.CreateStr("Say hello in a friendly way. Keep it very short."),
    Modalities = new List<ResponsesCreateRequestModality>
    {
        ResponsesCreateRequestModality.Text,
        ResponsesCreateRequestModality.Audio,
    },
});

// The output includes an output_audio item with base64 data and a transcript

mka1 llm responses create \
  -H 'X-On-Behalf-Of: <end-user-id>' \
  --body '{
    "model": "auto",
    "input": "Say hello in a friendly way. Keep it very short.",
    "modalities": ["text", "audio"]
  }'

curl https://apigw.mka1.com/api/v1/llm/responses \
  --request POST \
  --header 'Content-Type: application/json' \
  --header 'Authorization: Bearer <mka1-api-key>' \
  --header 'X-On-Behalf-Of: <end-user-id>' \
  --data '{
    "model": "auto",
    "input": "Say hello in a friendly way. Keep it very short.",
    "modalities": ["text", "audio"]
  }'

The response contains an output_audio item with the base64-encoded audio and a transcript of what was spoken:

{
  "status": "completed",
  "output": [
    {
      "type": "message",
      "role": "assistant",
      "content": [
        { "type": "output_text", "text": "Hello!" }
      ]
    },
    {
      "type": "output_audio",
      "id": "audio_460caf1079b34fa0b4aa74448dff4ea7",
      "data": "<Base64-encoded WAV audio data>",
      "transcript": "Hello!",
      "status": "completed"
    }
  ]
}

The data field contains the complete WAV file. The transcript field contains the text that was spoken, which is the reply’s output text. If speech synthesis fails, the response still completes, without an output_audio item.

Save audio to a file

const result = await sdk.llm.responses.create({
  responsesCreateRequest: {
    model: 'auto',
    input: 'Read this sentence aloud: The quick brown fox jumps over the lazy dog.',
    modalities: ['text', 'audio'],
  },
}) as ResponseObject;

// Find the audio output in the response
const audioItem = result.output.find((item) => item.type === 'output_audio');
if (audioItem) {
  const audioBuffer = Buffer.from(audioItem.data, 'base64');
  writeFileSync('output.wav', audioBuffer);
}

const response = await openai.post<{ output: Array<{ type: string; data?: string; transcript?: string }> }>('/responses', { body: {
  model: 'auto',
  input: 'Read this sentence aloud: The quick brown fox jumps over the lazy dog.',
  modalities: ['text', 'audio'],
  stream: false,
} });

const audioItem = response.output.find((item) => item.type === 'output_audio');
if (audioItem?.data) {
  const audioBuffer = Buffer.from(audioItem.data, 'base64');
  writeFileSync('output.wav', audioBuffer);
}

result = sdk.llm.responses.create(
    model="auto",
    input="Read this sentence aloud: The quick brown fox jumps over the lazy dog.",
    modalities=["text", "audio"],
)

# Find the audio output in the response
for item in result.output:
    if item.type == "output_audio":
        audio_bytes = base64.b64decode(item.data)
        with open("output.wav", "wb") as f:
            f.write(audio_bytes)

var result = await sdk.Llm.Responses.CreateAsync(new ResponsesCreateRequest()
{
    Model = "auto",
    Input = ResponsesCreateRequestInput.CreateStr(
        "Read this sentence aloud: The quick brown fox jumps over the lazy dog."),
    Modalities = new List<ResponsesCreateRequestModality>
    {
        ResponsesCreateRequestModality.Text,
        ResponsesCreateRequestModality.Audio,
    },
});

// Save the audio output to a file
// (iterate result.Output to find the output_audio item and decode its base64 data)

# Generate audio and extract the base64 data, then decode to a file
mka1 llm responses create \
  --body '{
    "model": "auto",
    "input": "Read this sentence aloud: The quick brown fox jumps over the lazy dog.",
    "modalities": ["text", "audio"]
  }' \
  --output-format json \
  --jq '.output[] | select(.type == "output_audio") | .data' | base64 -d > output.wav

# Generate audio and extract the base64 data, then decode to a file
curl -s https://apigw.mka1.com/api/v1/llm/responses \
  --request POST \
  --header 'Content-Type: application/json' \
  --header 'Authorization: Bearer <mka1-api-key>' \
  --data '{
    "model": "auto",
    "input": "Read this sentence aloud: The quick brown fox jumps over the lazy dog.",
    "modalities": ["text", "audio"]
  }' | jq -r '.output[] | select(.type == "output_audio") | .data' | base64 -d > output.wav

Supported languages

Audio responses speak the languages listed in Text to speech. A reply in any other language is spoken with the English voice. To choose the language yourself, or to get MP3 instead of WAV, send the reply text to Text to speech.

Get started

SDKs and CLI

Responses

Standalone APIs

Agents

Evals

Compute

Authentication

Production

Audio responses

Generate audio (text-to-speech)

Audio configuration

Save audio to a file

Supported languages

​Generate audio (text-to-speech)

​Audio configuration

​Save audio to a file

​Supported languages

Generate audio (text-to-speech)

Audio configuration

Save audio to a file

Supported languages