Text-only domain adaptation for end-to-end ASR using integrated text-to-mel-spectrogram generator
This addresses domain adaptation for ASR systems, allowing them to leverage text-only data from new domains, which is incremental but practical for real-world applications.
The paper tackles domain adaptation for end-to-end automatic speech recognition (ASR) by enabling training with text-only data, using an integrated text-to-mel-spectrogram generator with a GAN-based enhancer. The result is significantly improved ASR accuracy compared to speech-only training, with better adaptation quality and faster training than cascade TTS systems.
We propose an end-to-end Automatic Speech Recognition (ASR) system that can be trained on transcribed speech data, text-only data, or a mixture of both. The proposed model uses an integrated auxiliary block for text-based training. This block combines a non-autoregressive multi-speaker text-to-mel-spectrogram generator with a GAN-based enhancer to improve the spectrogram quality. The proposed system can generate a mel-spectrogram dynamically during training. It can be used to adapt the ASR model to a new domain by using text-only data from this domain. We demonstrate that the proposed training method significantly improves ASR accuracy compared to the system trained on transcribed speech only. It also surpasses cascade TTS systems with the vocoder in the adaptation quality and training speed.