Tacotron — End-to-End Text-to-Speech Model

Fact

An end-to-end model that turns text into speech spectrograms.

In Plain Words

Tacotron is like a tiny voice actor in a vending machine. Feed it words, and out drops a voice recipe.

It helped make text-to-speech more direct. You meet ideas like it in audiobooks and voice assistants.

Related Concepts

• Text-to-Speech
• Sequence-to-Sequence Model
• Attention Mechanism
• WaveNet

TTS
Tacotron is a key end-to-end model for neural text-to-speech.

Seq2Seq
Tacotron uses Seq2Seq to turn text sequences into spectrogram sequences.

Attention
Attention helps Tacotron match text to the sound being made.

WaveNet
Tacotron 2 often uses WaveNet to turn spectrograms into waveforms.