Audio8-tts-preview-0.6b: The Multilingual Text-to-Speech Model That Supports Speech Generation
Overview
Audio8-TTS-Preview-0.6b is a 0.6-billion-parameter multilingual text-to-speech model from Edge0 that supports ordinary speech generation and zero-shot voice cloning. Its main model contains 601,159,424 parameters, excluding the bundled codec. It uses a DualAR architecture inspired by Fish Audio S2 Pro: a slow autoregressive transformer predicts one semantic token per audio frame, while a fast autoregressive transformer predicts the codec codebooks conditioned on the slow transformer’s hidden state and preceding codebooks.
The model includes its tokenizer, processor, custom Hugging Face Transformers code, and 44.1 kHz neural audio codec, so it does not require a separate codec checkpoint. The packed text/audio context reaches 2,048 positions. The most important qualification is its Preview status: the recommended language set contains 11 languages, dialect coverage remains limited, and reference-audio quality and transcript accuracy affect cloning stability and speaker similarity. The standard implementation uses Python 3.10 or newer, PyTorch, torchaudio, Transformers with remote code, SoundFile, and...
Copyright of this story solely belongs to hackernoon.com. To see the full text click HERE
