Áudio na API Chat Completions

Se você já tem um aplicativo de texto que usa um LLM pelo endpoint de Chat Completions, talvez queira adicionar recursos de áudio. Por exemplo, se o seu aplicativo de chat aceita texto como entrada, você pode adicionar entrada e saída de áudio. Para isso, inclua audio no array modalities e use um modelo de áudio, como o gpt-audio-1.5.

Atualmente, a documentação da API Responses descreve entradas de texto e imagem com saídas em texto. Para implementar esse tipo de chat com áudio, use a API Chat Completions com um modelo compatível com áudio.

Gere uma resposta em áudio com voz natural a partir de um prompt

JavaScript

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28import { writeFileSync } from "node:fs";
import OpenAI from "openai";

const openai = new OpenAI();

// Generate an audio response to the given prompt
const response = await openai.chat.completions.create({
  model: "gpt-audio-1.5",
  modalities: ["text", "audio"],
  audio: { voice: "alloy", format: "wav" },
  messages: [
    {
      role: "user",
      content: "Is a golden retriever a good family dog?",
    },
  ],
  store: true,
});

// Inspect returned data
console.log(response.choices[0]);

// Write audio data to a file
writeFileSync(
  "dog.wav",
  Buffer.from(response.choices[0].message.audio.data, "base64"),
  { encoding: "utf-8" }
);

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17import base64
from openai import OpenAI

client = OpenAI()

completion = client.chat.completions.create(
    model="gpt-audio-1.5",
    modalities=["text", "audio"],
    audio={"voice": "alloy", "format": "wav"},
    messages=[{"role": "user", "content": "Is a golden retriever a good family dog?"}],
)

print(completion.choices[0])

wav_bytes = base64.b64decode(completion.choices[0].message.audio.data)
with open("dog.wav", "wb") as f:
    f.write(wav_bytes)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34package main

import (
	"context"
	"encoding/base64"
	"fmt"
	"os"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	response, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
		Model:      "gpt-audio-1.5",
		Modalities: []string{"text", "audio"},
		Audio: openai.ChatCompletionAudioParam{
			Voice:  openai.ChatCompletionAudioParamVoiceUnion{OfString: openai.String("alloy")},
			Format: openai.ChatCompletionAudioParamFormatWAV,
		},
		Messages: []openai.ChatCompletionMessageParamUnion{openai.UserMessage("Is a golden retriever a good family dog?")},
	})
	if err != nil {
		panic(err)
	}
	fmt.Println(response.Choices[0])
	audio, err := base64.StdEncoding.DecodeString(response.Choices[0].Message.Audio.Data)
	if err != nil {
		panic(err)
	}
	if err := os.WriteFile("dog.wav", audio, 0o600); err != nil {
		panic(err)
	}
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.chat.completions.ChatCompletionAudioParam;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.Base64;

ChatCompletionCreateParams params =
    ChatCompletionCreateParams.builder()
        .model("gpt-audio-1.5")
        .addUserMessage("Is a golden retriever a good family dog?")
        .addModality(ChatCompletionCreateParams.Modality.TEXT)
        .addModality(ChatCompletionCreateParams.Modality.AUDIO)
        .audio(
            ChatCompletionAudioParam.builder()
                .voice("alloy")
                .format(ChatCompletionAudioParam.Format.WAV)
                .build())
        .store(false)
        .build();

var message = client.chat().completions().create(params).choices().get(0).message();
var audio =
    message.audio().orElseThrow(() -> new IllegalStateException("No audio output returned"));
Files.write(Path.of("dog.wav"), Base64.getDecoder().decode(audio.data()));
message.content().ifPresent(System.out::println);

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25using OpenAI.Chat;
#pragma warning disable OPENAI001

string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ChatClient client = new("gpt-audio-1.5", key);

ChatCompletionOptions options = new()
{
    ResponseModalities = ChatResponseModalities.Text | ChatResponseModalities.Audio,
    AudioOptions = new(ChatOutputAudioVoice.Alloy, ChatOutputAudioFormat.Wav),
    StoredOutputEnabled = false,
};

ChatCompletion completion = await client.CompleteChatAsync(
    [new UserChatMessage("Is a golden retriever a good family dog?")],
    options
);

if (completion.OutputAudio is not ChatOutputAudio audio)
{
    throw new InvalidOperationException("No audio output was returned.");
}

Console.WriteLine(audio.Transcript);
await File.WriteAllBytesAsync("dog.wav", audio.AudioBytes.ToArray());

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22require "base64"
require "openai"

client = OpenAI::Client.new
completion = client.chat.completions.create(
  model: "gpt-audio-1.5",
  messages: [
    {
      role: :user,
      content: "Is a golden retriever a good family dog?"
    }
  ],
  modalities: [:text, :audio],
  audio: {
    voice: :alloy,
    format: :wav
  },
  store: true
)

audio = completion.choices.fetch(0).message.audio or raise "No audio returned"
File.binwrite("dog.wav", Base64.strict_decode64(audio.data))

1
2
3
4
5
6
7
8
9
10
11
12
13
14curl "https://api.openai.com/v1/chat/completions" \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer $OPENAI_API_KEY" \
    -d '{
      "model": "gpt-audio-1.5",
      "modalities": ["text", "audio"],
      "audio": { "voice": "alloy", "format": "wav" },
      "messages": [
        {
          "role": "user",
          "content": "Is a golden retriever a good family dog?"
        }
      ]
    }'

Use entradas de áudio como prompts para o modelo

JavaScript

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29import OpenAI from "openai";
const openai = new OpenAI();

// Fetch an audio file and convert it to a base64 string
const url = "https://cdn.openai.com/API/docs/audio/alloy.wav";
const audioResponse = await fetch(url);
const buffer = await audioResponse.arrayBuffer();
const base64str = Buffer.from(buffer).toString("base64");

const response = await openai.chat.completions.create({
  model: "gpt-audio-1.5",
  modalities: ["text", "audio"],
  audio: { voice: "alloy", format: "wav" },
  messages: [
    {
      role: "user",
      content: [
        { type: "text", text: "What is in this recording?" },
        {
          type: "input_audio",
          input_audio: { data: base64str, format: "wav" },
        },
      ],
    },
  ],
  store: false,
});

console.log(response.choices[0]);

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33import base64
import requests
from openai import OpenAI

client = OpenAI()

# Fetch the audio file and convert it to a base64 encoded string
url = "https://cdn.openai.com/API/docs/audio/alloy.wav"
response = requests.get(url)
response.raise_for_status()
wav_data = response.content
encoded_string = base64.b64encode(wav_data).decode("utf-8")

completion = client.chat.completions.create(
    model="gpt-audio-1.5",
    store=False,
    modalities=["text", "audio"],
    audio={"voice": "alloy", "format": "wav"},
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "What is in this recording?"},
                {
                    "type": "input_audio",
                    "input_audio": {"data": encoded_string, "format": "wav"},
                },
            ],
        },
    ],
)

print(completion.choices[0].message)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37package main

import (
	"context"
	"encoding/base64"
	"fmt"
	"os"

	"github.com/openai/openai-go/v3"
)

func main() {
	audio, err := os.ReadFile("fixtures/audio.wav")
	if err != nil {
		panic(err)
	}
	client := openai.NewClient()
	response, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
		Model:      "gpt-audio-1.5",
		Modalities: []string{"text", "audio"},
		Audio: openai.ChatCompletionAudioParam{
			Voice:  openai.ChatCompletionAudioParamVoiceUnion{OfString: openai.String("alloy")},
			Format: openai.ChatCompletionAudioParamFormatWAV,
		},
		Messages: []openai.ChatCompletionMessageParamUnion{openai.UserMessage([]openai.ChatCompletionContentPartUnionParam{
			openai.TextContentPart("What is in this recording?"),
			openai.InputAudioContentPart(openai.ChatCompletionContentPartInputAudioInputAudioParam{
				Data:   base64.StdEncoding.EncodeToString(audio),
				Format: "wav",
			}),
		})},
	})
	if err != nil {
		panic(err)
	}
	fmt.Println(response.Choices[0])
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.chat.completions.ChatCompletionAudioParam;
import com.openai.models.chat.completions.ChatCompletionContentPart;
import com.openai.models.chat.completions.ChatCompletionContentPartInputAudio;
import com.openai.models.chat.completions.ChatCompletionContentPartText;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import com.openai.models.chat.completions.ChatCompletionUserMessageParam;
import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.Base64;
import java.util.List;

String encodedAudio =
    Base64.getEncoder()
        .encodeToString(
            Files.readAllBytes(Path.of(System.getenv("OPENAI_EXAMPLE_AUDIO_PATH"))));

ChatCompletionCreateParams params =
    ChatCompletionCreateParams.builder()
        .model("gpt-audio-1.5")
        .addMessage(
            ChatCompletionUserMessageParam.builder()
                .contentOfArrayOfContentParts(
                    List.of(
                        ChatCompletionContentPart.ofText(
                            ChatCompletionContentPartText.builder()
                                .text("What is in this recording?")
                                .build()),
                        ChatCompletionContentPart.ofInputAudio(
                            ChatCompletionContentPartInputAudio.builder()
                                .inputAudio(
                                    ChatCompletionContentPartInputAudio.InputAudio.builder()
                                        .data(encodedAudio)
                                        .format(
                                            ChatCompletionContentPartInputAudio.InputAudio
                                                .Format.WAV)
                                        .build())
                                .build())))
                .build())
        .addModality(ChatCompletionCreateParams.Modality.TEXT)
        .addModality(ChatCompletionCreateParams.Modality.AUDIO)
        .audio(
            ChatCompletionAudioParam.builder()
                .voice("alloy")
                .format(ChatCompletionAudioParam.Format.WAV)
                .build())
        .store(false)
        .build();

client.chat().completions().create(params).choices().stream()
    .flatMap(choice -> choice.message().content().stream())
    .forEach(System.out::println);

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33using OpenAI.Chat;
#pragma warning disable OPENAI001

string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ChatClient client = new("gpt-audio-1.5", key);

BinaryData audio = BinaryData.FromBytes(
    await File.ReadAllBytesAsync("audio.wav")
);
UserChatMessage message = new(
    [
        ChatMessageContentPart.CreateTextPart("What is in this recording?"),
        ChatMessageContentPart.CreateInputAudioPart(
            audio,
            ChatInputAudioFormat.Wav
        ),
    ]
);
ChatCompletionOptions options = new()
{
    ResponseModalities = ChatResponseModalities.Text | ChatResponseModalities.Audio,
    AudioOptions = new(ChatOutputAudioVoice.Alloy, ChatOutputAudioFormat.Wav),
    StoredOutputEnabled = false,
};

ChatCompletion completion = await client.CompleteChatAsync([message], options);

if (completion.OutputAudio is not ChatOutputAudio audioOutput)
{
    throw new InvalidOperationException("No audio output was returned.");
}

Console.WriteLine(audioOutput.Transcript);

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35require "base64"
require "openai"

client = OpenAI::Client.new
audio = Base64.strict_encode64(File.binread("audio.wav"))
completion = client.chat.completions.create(
  model: "gpt-audio-1.5",
  messages: [
    {
      role: :user,
      content: [
        {
          type: :text,
          text: "What is in this recording?"
        },
        {
          type: :input_audio,
          input_audio: {
            data: audio,
            format: :wav
          }
        }
      ]
    }
  ],
  modalities: [:text, :audio],
  audio: {
    voice: :alloy,
    format: :wav
  },
  store: false
)

message = completion.choices.fetch(0).message
puts(message.audio&.transcript || message.content)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23curl "https://api.openai.com/v1/chat/completions" \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer $OPENAI_API_KEY" \
    -d '{
      "model": "gpt-audio-1.5",
      "modalities": ["text", "audio"],
      "audio": { "voice": "alloy", "format": "wav" },
      "messages": [
        {
          "role": "user",
          "content": [
            { "type": "text", "text": "What is in this recording?" },
            { 
              "type": "input_audio", 
              "input_audio": { 
                "data": "<base64 bytes here>", 
                "format": "wav" 
              }
            }
          ]
        }
      ]
    }'

Sugestões

Primeiros passos

Conceitos fundamentais

SDKs e CLI

Provedores de nuvem

Recursos

APIs legadas

Escolha um modelo

Texto e código

Como escrever prompts

Raciocínio

Imagens

Tempo real e áudio

Modelos especializados

API de agentes

SDK de agentes

ChatKit

Busca e recuperação de informações

Conecte ferramentas e dados

Crie fluxos de trabalho com ferramentas

Computador e código

Mídia

GPT-Live

Realtime API

Desenvolver aplicações de voz

Conexões

Processamento de áudio

Coloque em produção

Desempenho e qualidade

Avaliação

Custo e taxa de processamento

Segurança e governança

Infraestrutura e acesso

Operações

Conceitos fundamentais

Planejar

Desenvolver

Testar e publicar

Especificações de conversão

Guias

Recursos

Identidade

Uso do plano do ChatGPT

Guias

Envio de arquivos

API

Mensuração

API para anunciantes

Referência da API

Primeiros passos

Primeiros passos

Fundamentos

Explorar

Disponível em

Versões

Fluxos de trabalho

Capacidades

Sites

dots

ChatGPT Space

Referência

Personalização

Arquivo de configuração

Configuração de agentes

Ampliar os recursos do ChatGPT e do Codex

Linux

Windows

Fluxos de trabalho de desenvolvimento

Ampliar recursos e automatizar tarefas

Ambientes

Codex Cloud

Desenvolver com o Codex

Integrações de terceiros

Referência

Permissões

Codex Security

Segurança cibernética

Primeiros passos

Configuração de recursos

Identidade e acesso