Transcribe Audio
Transcribe audio files into the input language using the Whisper model.
require "openai"
require "pathname"
client = OpenAI::Client.new
response = client.audio.transcriptions.create(
file: Pathname("audio.mp3"),
model: "gpt-4o-transcribe"
)
puts response.text
The file parameter accepts a Pathname, StringIO, IO, or raw file contents.
Transcription with Options
Customize transcription with additional parameters like language, prompt, and response format.
response = client.audio.transcriptions.create(
file: Pathname("meeting.mp3"),
model: "gpt-4o-transcribe",
language: "en",
prompt: "This is a meeting about product development",
response_format: "verbose_json",
temperature: 0.2
)
# Verbose response includes timestamps and other metadata
puts "Text: #{response.text}"
puts "Language: #{response.language}"
puts "Duration: #{response.duration}"
JSON (default)
Verbose JSON
Diarized JSON
Plain Text
response = client.audio.transcriptions.create(
file: Pathname("audio.mp3"),
model: "gpt-4o-transcribe",
response_format: "json"
)
# Returns basic transcription text
puts response.text
response = client.audio.transcriptions.create(
file: Pathname("audio.mp3"),
model: "gpt-4o-transcribe",
response_format: "verbose_json"
)
# Returns detailed information
puts response.text
puts response.language
puts response.duration
response.segments.each do |segment|
puts "#{segment.start} - #{segment.end}: #{segment.text}"
end
response = client.audio.transcriptions.create(
file: Pathname("audio.mp3"),
model: "gpt-4o-transcribe",
response_format: "diarized_json"
)
# Returns speaker-separated transcription
response.segments.each do |segment|
puts "Speaker #{segment.speaker}: #{segment.text}"
end
response = client.audio.transcriptions.create(
file: Pathname("audio.mp3"),
model: "gpt-4o-transcribe",
response_format: "text"
)
# Returns plain text string
puts response
Translate Audio to English
Translate audio files from any supported language into English.
response = client.audio.translations.create(
file: Pathname("french_audio.mp3"),
model: "whisper-1"
)
puts response.text # English translation
Translation with Prompt
Provide context to guide the translation style or continue from previous audio.
response = client.audio.translations.create(
file: Pathname("spanish_interview.mp3"),
model: "whisper-1",
prompt: "This is an interview with a software engineer discussing AI",
response_format: "verbose_json"
)
puts "Translation: #{response.text}"
puts "Original language: #{response.language}"
Translation always converts audio to English, regardless of the source language. Use transcription if you want to keep the original language.
- FLAC
- MP3
- MP4
- MPEG
- MPGA
- M4A
- OGG
- WAV
- WEBM
# Any supported format works
formats = ["audio.flac", "audio.mp3", "audio.wav", "audio.m4a"]
formats.each do |file|
response = client.audio.transcriptions.create(
file: Pathname(file),
model: "gpt-4o-transcribe"
)
puts "#{file}: #{response.text}"
end
Using StringIO
Transcribe audio from memory without writing to disk.
require "stringio"
# Read audio file into memory
audio_data = File.read("audio.mp3")
audio_io = StringIO.new(audio_data)
response = client.audio.transcriptions.create(
file: audio_io,
model: "gpt-4o-transcribe"
)
puts response.text