Speech-to-text (STT), also called automatic speech recognition (ASR), is the technology that converts a stream of audio containing human speech into a written transcript. Modern systems are end-to-end neural models—most prominently OpenAI's Whisper—trained on hundreds of thousands of hours of multilingual audio, enabling them to handle accents, background noise, overlapping speech, and technical vocabulary with high accuracy. Outputs typically include punctuation, timestamps, and optionally speaker labels produced by a separate diarization layer.

Think of it like a court stenographer who has internalized the acoustic patterns of every regional accent and technical vocabulary. Rather than learning shorthand symbols, the model has learned a statistical mapping from raw sound waves to sequences of words, refined across millions of hours of diverse recorded speech.

Using Whisper in Python takes just a few lines: import whisper; model = whisper.load_model("large-v3"); result = model.transcribe("interview.mp3", language="en"); print(result["text"]). For real-time applications, streaming variants process audio in chunks as small as 100 ms, feeding partial transcripts to a downstream application before the speaker has finished.

STT output is raw text that feeds directly into Tokenization pipelines for LLM processing, and the acoustic encoder inside modern ASR systems is itself a Transformer that uses Attention over mel-spectrogram frames; STT pairs naturally with Multimodal systems. You will encounter it in voice assistants, meeting transcription tools, accessibility software, subtitle generation pipelines, and any application that needs to accept spoken input.