Audio API

OpenAI-compatible audio surface. The SDK accepts the same inputs the OpenAI Audio API does — STT, translation, and TTS — but routes them through the Console gateway.

client.audio.transcriptions.create(params)

Speech-to-text. Accepts a base64 string, a structured { kind: 'blob' | 'buffer' | 'base64' } source, or an inline { data, fileName } object.

import { readFileSync } from 'node:fs';
import { ConsoleClient } from '@cognipeer/console-sdk';

const client = new ConsoleClient({
  apiKey: process.env.COGNIPEER_API_KEY!,
  baseURL: process.env.COGNIPEER_BASE_URL,
});

const audioBuf = readFileSync('voice.mp3');

const transcript = await client.audio.transcriptions.create({
  model: 'whisper-1',
  audio: {
    kind: 'buffer',
    data: audioBuf,
    fileName: 'voice.mp3',
    contentType: 'audio/mpeg',
  },
  response_format: 'verbose_json',
  timestamp_granularities: ['segment'],
});

console.log(transcript.text);

Parameters

Name	Type	Description
model	string	Model key (e.g. whisper-1).
audio	string | AudioFileInput | AudioFileSource	Audio payload. Strings are treated as base64.
language	string	Source language.
prompt	string	Bias prompt.
response_format	'json' | 'text' | 'srt' | 'verbose_json' | 'vtt'	Response shape.
temperature	number	Sampling temperature.
timestamp_granularities	Array<'word' | 'segment'>	Required granularities when using verbose_json.

Audio input shapes

type AudioFileSource =
  | { kind: 'base64'; data: string; fileName?: string; contentType?: string }
  | { kind: 'blob';   blob: Blob; fileName?: string }
  | { kind: 'buffer'; data: Uint8Array | ArrayBuffer; fileName?: string; contentType?: string };

interface AudioFileInput {
  data: string;        // raw base64
  fileName?: string;
  contentType?: string;
}

When you pass a Blob/Buffer/AudioFileSource, the SDK sends the request as multipart/form-data. Plain base64 strings are sent as JSON, which is useful for serverless runtimes that can't construct multipart bodies.

client.audio.translations.create(params)

Same shape as transcriptions.create, but translates the audio to English on the server.

const en = await client.audio.translations.create({
  model: 'whisper-1',
  audio: { kind: 'buffer', data: readFileSync('greeting.tr.mp3') },
});

client.audio.speech.create(params)

Synthesize speech. Returns the raw audio bytes (Uint8Array) plus the response Content-Type header.

import { writeFileSync } from 'node:fs';

const speech = await client.audio.speech.create({
  model: 'tts-1',
  input: 'Hello world',
  voice: 'alloy',
  response_format: 'mp3',
});

writeFileSync('hello.mp3', speech.audio);
console.log(speech.contentType); // audio/mpeg

Parameters

Name	Type	Description
model	string	TTS model key (e.g. tts-1).
input	string	Text to synthesize.
voice	string	Voice identifier.
response_format	'mp3' | 'opus' | 'aac' | 'flac' | 'wav' | 'pcm'	Output format.
speed	number	Playback speed multiplier.
instructions	string	Optional style instructions.

Response shape

interface AudioSpeechResponse {
  audio: Uint8Array;
  contentType: string;
  requestId?: string;
}