Speech Recognition

Definisi

Speech recognition (ASR - Automatic Speech Recognition) adalah teknologi yang mengkonversi audio suara menjadi teks.

Sejarah

• 1952 — Audrey (Bell Labs) — recognizer digit
• 1971 — Harpy (CMU) — 1000 kata
• 1980-an — Hidden Markov Models (HMM)
• 1990-an — commercial dictation (Dragon NaturallySpeaking)
• 2000-an — large vocabulary continuous speech recognition (LVCSR)
• 2010-an — Deep Neural Networks (DNN)
• 2016+ — end-to-end deep learning (DeepSpeech, wav2vec, Conformer)
• 2022 — Whisper (OpenAI) — multilingual, robust
• 2024+ — multimodal speech (audio + visual)

Sistem Modern

• Whisper (OpenAI) — multilingual 99 bahasa
• Google Cloud Speech-to-Text
• AWS Transcribe
• Azure Speech
• AssemblyAI, Deepgram
• Apple Dictation

Aplikasi

• Voice assistants (Siri, Google Assistant, Alexa)
• Transcription (Otter, Fireflies, Rev)
• Subtitle/captioning
• Medical dictation
• Customer service analytics
• Real-time translation

Akurasi

Pada clean English, modern systems mencapai < 5% WER (Word Error Rate) — mendekati manusia. Pada noisy/accented, masih challenging.