Principle:Openai Openai python Speech Transcription
| Knowledge Sources | |
|---|---|
| Domains | Audio, Speech_Recognition |
| Last Updated | 2026-02-15 00:00 GMT |
Overview
An automatic speech recognition technique that converts audio input into text with support for multiple languages, timestamps, and speaker diarization.
Description
Speech transcription (ASR) converts spoken audio into written text. Modern models like Whisper and GPT-4o Transcribe support multiple languages, optional word-level timestamps, speaker diarization, streaming transcription, and various output formats (plain text, JSON, SRT subtitles, VTT). Language hints and prompt context can improve accuracy for domain-specific content.
Usage
Use this principle when you need to convert audio recordings or live audio streams to text. Applications include meeting transcription, subtitle generation, voice command processing, and accessibility features. Choose streaming mode for real-time transcription of live audio.
Theoretical Basis
Transcription applies a Sequence-to-Sequence model to audio:
# Transcription flow
text = transcribe(
audio_file=audio,
model=asr_model, # whisper-1, gpt-4o-transcribe
language="en", # Optional language hint
response_format="json", # Output format
timestamp_granularity=["word", "segment"] # Timing info
)
# Streaming variant
for event in transcribe_stream(audio_file, model):
process_partial(event.text)
The model processes audio in chunks, applying learned acoustic and language models to produce text with optional timing alignment.