Utterance

Observation noise: 0.15 Sequence length: 24 frames

Viterbi playback

Playback speed: 1.0×

Decode progress

Phase—

Column0 / 24

Best running log-prob—

Result

Decoding accuracy—

Log-likelihood (decoded path)—

True phoneme path—

Decoded phoneme path—

The top strip shows the discretized acoustic observation symbol at each time frame. The trellis below has one row per hidden phoneme state and one column per time frame; brightness encodes each cell's running max log-probability under the real Viterbi dynamic program. The white outline marks the true (generating) hidden path; the solid violet line traces the backpointers of the winning decoded path once the forward pass and backtrace complete.