Search Results for "frame"

...Unlike traditional TTS systems, it excels in scalability, speaker consistency, and natural turn-taking for up to 90 minutes of continuous speech with as many as four distinct speakers. A key innovation is its use of continuous acoustic and semantic speech tokenizers operating at an ultra-low frame rate of 7.5 Hz, enabling high audio fidelity with efficient processing of long sequences. The model integrates a Qwen2.5-based large language model with a diffusion head to produce realistic acoustic details and capture conversational context. Training involved curriculum learning with increasing sequence lengths up to 65K tokens, allowing VibeVoice to handle very long dialogues effectively. ...

...It takes inspiration from architectures like FastSpeech, FastSpeech 2, FastPitch, and Transformer TTS, and extends them with its own aligner and forward models. The system separates alignment learning and acoustic modeling: an autoregressive Transformer is used as an aligner to extract phoneme-to-frame durations, while a non-autoregressive “ForwardTransformer” generates mel-spectrograms conditioned on text and durations. This design addresses common autoregressive issues such as repetition, skipped words, and unstable attention, and results in robust, fast synthesis where all frames are predicted in parallel. The repository ships with tooling to build datasets (especially LJSpeech) and create training data, plus scripts to train both the aligner and the TTS model, monitor training with TensorBoard, and resume or reset training runs.