DataTerminology.comNEW
Also known as: ASR
Automatically transcribing spoken audio into written text using acoustic and language modeling techniques.
DESCRIPTION
Speech-to-text systems map audio waveforms to text via feature extraction, acoustic modeling, and decoding with language models. Modern systems use transformers and self-supervised pretraining. Applications include captioning, voice assistants, and call analytics. Challenges involve accents, noise, domain adaptation, and latency. On-device and streaming ASR address privacy and real-time needs.
ALTERNATIVE DEFINITIONS
Wikipedia
Speech recognition is an interdisciplinary subfield of computer science and computational linguistics that develops methodologies and technologies that enable the recognition and translation of spoken language into text by computers.
Frames ASR as a discipline enabling recognition and translation into text.
Google Cloud
Speech-to-Text enables developers to convert audio to text by applying powerful neural network models.
Emphasizes developer-facing API and neural network modeling.
RELATED TERMS
Speech Recognition Language Model Acoustic Model Connectionist Temporal Classification Attention Mechanism Speaker Diarization Endpointing Noise Reduction Keyword Spotting
COMPETING TERMS
KNOWLEDGE AREAS
TARGET AUDIENCE
AI Engineer Data Scientist MLOps Engineer AI Product Manager
EXTERNAL SOURCES
- Wikipedia - Speech recognition
- Google Cloud - Managed ASR service overview
- Microsoft - Azure speech-to-text docs
- arXiv - Wav2Vec 2.0 self-supervised ASR
TOP QUESTIONS
What is Speech-to-Text used for?
It transcribes spoken audio to text for captions, search, voice assistants, and analytics. It enables accessible and indexable content.
Category: Overview
How do modern Speech-to-Text systems work?
They use neural acoustic models and language models with beam search decoding. Self-supervised pretraining improves robustness.
Category: Concepts
What are common challenges in Speech-to-Text?
Accents, noisy environments, domain-specific jargon, and real-time latency complicate accuracy and usability.
Category: Mistakes
How can I improve Speech-to-Text accuracy?
Use domain adaptation, custom vocabularies, noise suppression, and diarization. Evaluate with representative audio.
Category: HowTo
What are the risks of using Speech-to-Text?
Privacy issues with sensitive audio, bias across dialects, and transcription errors affecting downstream decisions are key risks.
Category: Risks
How does Speech-to-Text relate to Text-to-Speech?
They are inverse tasks: ASR converts audio to text; TTS synthesizes speech from text. Both often use transformers.
Category: Comparison
What latency can streaming ASR achieve?
Modern systems achieve sub-second partial hypotheses with finalization within seconds, depending on bandwidth and model size.
Category: Opinion
Where should I start with Speech-to-Text?
Try managed APIs for quick prototypes, then explore open-source toolkits and self-supervised models for customization.
Category: Learning
What metrics evaluate Speech-to-Text?
Word error rate (WER), character error rate (CER), and latency are common metrics; domain-specific accuracy matters too.
Category: Concepts