Research map / Label-efficient and robust learning / Recognition and long-tail
Almost Unsupervised Text to Speech and Automatic Speech Recognition
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use denoising auto-encoding on unpaired speech and text (ablation) | Pair-200 · LJSpeech test set | PER (ASR): 72.3% -> 52.0% | Helped |
| Add dual transformation between TTS and ASR (ablation) | Pair-200 + DAE · LJSpeech test set | PER (ASR): 52.0% -> 15.3% | Helped |
| Add dual transformation between TTS and ASR (ablation) | Pair-200 + DAE · LJSpeech | MOS (TTS): Null -> 2.11 | Helped |
| Add bidirectional sequence modeling (ablation) | Pair-200 + DAE + DT · LJSpeech | MOS (TTS): 2.11 -> 2.51 | Helped |
| Add bidirectional sequence modeling (ablation) | Pair-200 + DAE + DT · LJSpeech test set | PER (ASR): 15.3% -> 11.7% | Helped |
| Use the complete method with unpaired data and 200 paired examples rather than paired data only | Pair-200 · LJSpeech | MOS (TTS): Null -> 2.68 | Helped |
| Use the complete method with unpaired data and 200 paired examples rather than paired data only | Pair-200 · LJSpeech test set | PER (ASR): 72.3% -> 11.7% | Helped |
| Increase paired examples from 100 to 200 (ablation) | Our Method · LJSpeech test set | PER (ASR): 64.2% -> 11.7% | Helped |
| Increase paired examples from 200 to 300 (ablation) | Our Method · LJSpeech test set | PER (ASR): 11.7% -> 8.4% | Helped |
| Increase paired examples from 300 to 400 (ablation) | Our Method · LJSpeech test set | PER (ASR): 8.4% -> 5.2% | Helped |
| Increase paired examples from 400 to 500 (ablation) | Our Method · LJSpeech test set | PER (ASR): 5.2% -> 4.4% | Helped |
| Increase paired examples from 100 to 500 (ablation) | Our Method · LJSpeech | MOS (TTS): Null -> 2.78 | Helped |
Closest papers by meaning
- MaskGCT: Zero-Shot Text-to-Speech with Masked Generative Codec Transformer (ICLR 2025 · similarity 0.88)
- Speech Recognition Meets Large Language Model: Benchmarking, Models, and Exploration (AAAI 2025 · similarity 0.87)
- Scaling Speech-Text Pre-training with Synthetic Interleaved Data (ICLR 2025 · similarity 0.87)
- Self-Taught Recognizer: Toward Unsupervised Adaptation for Speech Foundation Models (NeurIPS 2024 · similarity 0.87)
- CLaM-TTS: Improving Neural Codec Language Model for Zero-Shot Text-to-Speech (ICLR 2024 · similarity 0.87)
- DELTA-TTS: Adapting Autoregressive Model into a Diffusion Language Model for Text-to-Speech (NeurIPS 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.