TokenSplit: Using Discrete Speech Representations for Direct, Refined, and Transcript-Conditioned Speech Separation and Recognition
This work addresses speech separation and recognition for applications in audio processing, but it is incremental as it builds on existing methods with a hybrid approach.
The paper tackles speech separation by introducing TokenSplit, a model that uses discrete token sequences to separate and transcribe multiple speech sources, achieving excellent performance in separation tasks with or without transcript conditioning, as validated by objective metrics and subjective MUSHRA tests.
We present TokenSplit, a speech separation model that acts on discrete token sequences. The model is trained on multiple tasks simultaneously: separate and transcribe each speech source, and generate speech from text. The model operates on transcripts and audio token sequences and achieves multiple tasks through masking of inputs. The model is a sequence-to-sequence encoder-decoder model that uses the Transformer architecture. We also present a "refinement" version of the model that predicts enhanced audio tokens from the audio tokens of speech separated by a conventional separation model. Using both objective metrics and subjective MUSHRA listening tests, we show that our model achieves excellent performance in terms of separation, both with or without transcript conditioning. We also measure the automatic speech recognition (ASR) performance and provide audio samples of speech synthesis to demonstrate the additional utility of our model.