Skip to content

BASE TTS

A 1-billion-parameter text-to-speech model from Amazon

Visit official site

What BASE TTS is

The model is described as a large-scale text-to-speech system from Amazon that converts text into speech codes with a 1-billion-parameter autoregressive transformer, then produces waveforms through a convolutional decoder.

What you can do with it

  • Convert text to speech through the model pipeline
  • Turn speech codes into waveforms with a convolutional decoder
  • Apply phoneme disentanglement and compression in speech encoding
  • Handle natural intonation in complex sentences as model size grows

Application scenarios

  • Convert input text into lifelike speech
  • Generate narration for audiobooks automatically
  • Give a voice assistant a more natural intonation
  • Read written text aloud for people with visual impairments

Model notes

  • An autoregressive transformer with 1 billion parameters is described
  • Training uses more than 100000 hours of public speech data
  • The source calls the result a new state in speech naturalness
  • Novel speech encoding covers phoneme disentanglement and compression

Pros & cons

✓ What we like

  • Phoneme disentanglement and compression are listed as speech encoding techniques
  • Natural intonation for complex sentences grows with model size

! What to watch out for

  • The model turns text into speech codes before waveform generation
  • Training draws on more than 100000 hours of public speech data

FAQ

Who developed the model?

The material states the model was developed by Amazon.

Which areas are listed for it?

The listed areas are speech synthesis, voice assistants and audiobook generation.

Last reviewed: 2026-11-15

More AI audio tools tools

View all →

How we review