Class: LLM::OpenAI::Audio

Inherits:

Object

• Object
• LLM::OpenAI::Audio

Defined in:

lib/llm/providers/openai/audio.rb

Overview

The LLM::OpenAI::Audio class provides an audio object for interacting with OpenAI’s audio API.

Examples:

llm = LLM.openai(ENV["KEY"])
res = llm.audio.create_speech(input: "A dog on a rocket to the moon")
IO.copy_stream res.audio, "rocket.mp3"

Instance Method Summary

• #initialize(provider) ⇒ LLM::OpenAI::Responses constructor

  Returns a new Audio object.

• #create_speech(input:, voice: "alloy", model: "gpt-4o-mini-tts", response_format: "mp3", **params) ⇒ LLM::Response::Audio

  Create an audio track.

• #create_transcription(file:, model: "whisper-1", **params) ⇒ LLM::Response::AudioTranscription

  Create an audio transcription.

• #create_translation(file:, model: "whisper-1", **params) ⇒ LLM::Response::AudioTranslation

  Create an audio translation (in English).

Constructor Details

#initialize(provider) ⇒ LLM::OpenAI::Responses

Returns a new Audio object

Parameters:

• provider (LLM::Provider)

# File 'lib/llm/providers/openai/audio.rb', line 16

def initialize(provider)
  @provider = provider
end

Instance Method Details

#create_speech(input:, voice: "alloy", model: "gpt-4o-mini-tts", response_format: "mp3", **params) ⇒ LLM::Response::Audio

Create an audio track

Examples:

llm = LLM.openai(ENV["KEY"])
res = llm.images.create_speech(input: "A dog on a rocket to the moon")
File.binwrite("rocket.mp3", res.audio.string)

Parameters:

• input (String) —

  The text input

• voice (String) (defaults to: "alloy") —

  The voice to use

• model (String) (defaults to: "gpt-4o-mini-tts") —

  The model to use

• response_format (String) (defaults to: "mp3") —

  The response format

• params (Hash) —

  Other parameters (see OpenAI docs)

Returns:

• (LLM::Response::Audio)

Raises:

• (LLM::Error::Unauthorized) —

  When authentication fails

• (LLM::Error::RateLimit) —

  When the rate limit is exceeded

• (LLM::Error::ResponseError) —

  When any other unsuccessful status code is returned

• (SystemCallError) —

  When there is a network error at the operating system level

See Also:

• OpenAI docs

# File 'lib/llm/providers/openai/audio.rb', line 34

def create_speech(input:, voice: "alloy", model: "gpt-4o-mini-tts", response_format: "mp3", **params)
  req = Net::HTTP::Post.new("/v1/audio/speech", headers)
  req.body = JSON.dump({input:, voice:, model:, response_format:}.merge!(params))
  io = StringIO.new("".b)
  res = request(http, req) { _1.read_body { |chunk| io << chunk } }
  LLM::Response::Audio.new(res).tap { _1.audio = io }
end

#create_transcription(file:, model: "whisper-1", **params) ⇒ LLM::Response::AudioTranscription

Create an audio transcription

Examples:

llm = LLM.openai(ENV["KEY"])
res = llm.audio.create_transcription(file: "/audio/rocket.mp3")
res.text # => "A dog on a rocket to the moon"

Parameters:

• file (String, LLM::File) —

  The input audio

• model (String) (defaults to: "whisper-1") —

  The model to use

• params (Hash) —

  Other parameters (see OpenAI docs)

Returns:

• (LLM::Response::AudioTranscription)

Raises:

• (LLM::Error::Unauthorized) —

  When authentication fails

• (LLM::Error::RateLimit) —

  When the rate limit is exceeded

• (LLM::Error::ResponseError) —

  When any other unsuccessful status code is returned

• (SystemCallError) —

  When there is a network error at the operating system level

See Also:

• OpenAI docs

# File 'lib/llm/providers/openai/audio.rb', line 54

def create_transcription(file:, model: "whisper-1", **params)
  multi = LLM::Multipart.new(params.merge!(file: LLM.File(file), model:))
  req = Net::HTTP::Post.new("/v1/audio/transcriptions", headers)
  req["content-type"] = multi.content_type
  set_body_stream(req, multi.body)
  res = request(http, req)
  LLM::Response::AudioTranscription.new(res).tap { _1.text = _1.body["text"] }
end

#create_translation(file:, model: "whisper-1", **params) ⇒ LLM::Response::AudioTranslation

Create an audio translation (in English)

Examples:

# Arabic => English
llm = LLM.openai(ENV["KEY"])
res = llm.audio.create_translation(file: "/audio/bismillah.mp3")
res.text # => "In the name of Allah, the Beneficent, the Merciful."

Parameters:

• file (LLM::File) —

  The input audio

• model (String) (defaults to: "whisper-1") —

  The model to use

• params (Hash) —

  Other parameters (see OpenAI docs)

Returns:

• (LLM::Response::AudioTranslation)

Raises:

• (LLM::Error::Unauthorized) —

  When authentication fails

• (LLM::Error::RateLimit) —

  When the rate limit is exceeded

• (LLM::Error::ResponseError) —

  When any other unsuccessful status code is returned

• (SystemCallError) —

  When there is a network error at the operating system level

See Also:

• OpenAI docs

# File 'lib/llm/providers/openai/audio.rb', line 76

def create_translation(file:, model: "whisper-1", **params)
  multi = LLM::Multipart.new(params.merge!(file: LLM.File(file), model:))
  req = Net::HTTP::Post.new("/v1/audio/translations", headers)
  req["content-type"] = multi.content_type
  set_body_stream(req, multi.body)
  res = request(http, req)
  LLM::Response::AudioTranslation.new(res).tap { _1.text = _1.body["text"] }
end