ORCIDGitHubLinkedIn
RSS Curriculum Vitae TR

Connectionist Temporal Classification

Turkish equivalent: Bağlantıcı zamansal sınıflandırmaDomain: Speech Processing

A sequence-training objective that marginalizes over possible monotonic alignments between input frames and output labels using a blank symbol and collapse rules.

Speech Processing Context

Connectionist Temporal Classification trains sequence models when input and output lengths differ and explicit frame-level alignment is unavailable. It introduces a blank symbol and collapses repeated labels and blanks to obtain the final sequence, making it useful for monotonic alignment tasks such as ASR and OCR.

Modeling Boundary

CTC is not a language model and its common formulation assumes conditional independence between output labels at different time steps given the input. Beam search can combine CTC scores with an external language model during decoding.

Direct source: The primary paper or official specification for Connectionist Temporal Classification is linked here for verification.

Updated:

Related concepts

Speech Processing

Related technical publications

Publications whose title or summary directly references this concept.

Automatic Speech Recognition

Automatic speech recognition from isolated-word, word-spotting and continuous-speech tasks through DTW, HMM/WFST and n-gram systems to CTC, RNN-T, Transformer and Conformer models.