Voxtlm: unified decoder-only models for consolidating speech recognition/synthesis and speech/text continuation tasks
This work addresses the need for integrated models in speech and text domains, offering a unified approach that is incremental by combining existing methods with multitask learning.
The paper tackled the problem of unifying speech and text processing tasks by proposing VoxtLM, a decoder-only language model that performs speech recognition, synthesis, text generation, and speech continuation, resulting in significant improvements such as speech intelligibility increasing from 28.9 to 5.6 and objective quality from 2.68 to 3.90.
We propose a decoder-only language model, VoxtLM, that can perform four tasks: speech recognition, speech synthesis, text generation, and speech continuation. VoxtLM integrates text vocabulary with discrete speech tokens from self-supervised speech features and uses special tokens to enable multitask learning. Compared to a single-task model, VoxtLM exhibits a significant improvement in speech synthesis, with improvements in both speech intelligibility from 28.9 to 5.6 and objective quality from 2.68 to 3.90. VoxtLM also improves speech generation and speech recognition performance over the single-task counterpart. Further, VoxtLM is trained with publicly available data and training recipes and model checkpoints are open-sourced to make fully reproducible work.