Towards Transfer Learning for End-to-End Speech Synthesis from Deep Pre-Trained Language Models
원문 초록
Modern text-to-speech (TTS) systems are able to generate audio that sounds almost as natural as human speech. However, the bar of developing high-quality TTS systems remains high since a sizable set of studio-quality <text, audio> pairs is usually required. Compared to commercial data used to develop state-of-the-art systems, publicly available data are usually worse in terms of both quality and size. Audio generated by TTS systems trained on publicly available data tends to not only sound less natural, but also exhibits more background noise. In this work, we aim to lower TTS systems' reliance on high-quality data by providing them the textual knowledge extracted by deep pre-trained language models during training. In particular, we investigate the use of BERT to assist the training of Tacotron-2, a state of the art TTS consisting of an encoder and an attention-based decoder. BERT representations learned from large amounts of unlabeled text data are shown to contain very rich semantic and syntactic information about the input text, and have potential to be leveraged by a TTS system to compensate the lack of high-quality data. We incorporate BERT as a parallel branch to the Tacotron-2 encoder with its own attention head. For an input text, it is simultaneously passed into BERT and the Tacotron-2 encoder. The representations extracted by the two branches are concatenated and then fed to the decoder. As a preliminary study, although we have not found incorporating BERT into Tacotron-2 generates more natural or cleaner speech at a human-perceivable level, we observe improvements in other aspects such as the model is being significantly better at knowing when to stop decoding such that there is much less babbling at the end of the synthesized audio and faster convergence during training.
함께 살펴볼 논문
공통 태그를 근거로 연결했습니다. 실제 인용 관계나 논문의 우수성을 뜻하지 않습니다.
Effective Deep Learning Models for Automatic Diacritization of Arabic Text
Bts-e: Audio deepfake detection using breathing-talking-silence encoder
Mlphon: A Multifunctional Grapheme-Phoneme Conversion Tool Using Finite State Transducers
Multi-Stage Deep Transfer Learning for EmIoT-enabled Human-Computer Interaction
Wav2KWS: Transfer Learning from Speech Representations for Keyword Spotting
사용자 생성 텍스트의 음소화: 벤치마크, 분류 체계 및 조합적 접근
Phonemizing User-Generated Text: A Benchmark, Taxonomy, and Compositional Approach
텍스트 음성 변환 시스템은 ppl, imo와 같은 사용자 생성 텍스트(UGT)를 점점 더 많이 처리하며, 이러한 텍스트의 발음은 표면 형태가 아니라 정규(canonical) 형태로부터 추론해야 한다. 우리는 영어, 베트남어, 한국어 UGT를 위한 최초의 grapheme-to-phoneme(G2P) 벤치마크 UGTPhon과, 세밀한 진단을 위한 추론 기반 분류 체계를 제안한다. 기존 G2P 모델과 최신 프론티어 LLM은 체계적인 정규-비정규 성능 격차를 보이며, 최대 66.8 PER points에 이른다. 벤치마크 기준선으로, exact-match lookup과 단계적 디코딩(staged decoding)을 통해 정규 형태 증거를 통합하는 단순한 조합적 G2P 접근법을 제안한다. 매칭된 ByT5 및 Qwen2.5-0.5B 백본 전반에서 명시적 정규 형태 모델링은 비정규 G2P 오류를 일관되게 줄인다. 0.5B 변형은 훨씬 더 큰 퓨샷(few-shot) 프론티어 LLM과도 경쟁력 있는 성능을 보이며, UGT 음소화를 위해 정규 형태 추론을 명시적으로 모델링하는 이점을 부각한다.
토론
이 논문과 연결된 포럼 글이 아직 없습니다.