Research map / Probabilistic and scientific modeling / Molecules and proteins
Sequence to Multi-Sequence Learning via Conditional Chain Mapping for Mixture Signals
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 |
|---|---|---|---|
| Add conditional chain (single extra LSTM layer) to TasNet | TasNet · WSJ0-2mix | SI-SNRi: 15.4 -> 15.6 | Helped |
| Add conditional chain to TasNet, 3-speaker | TasNet · WSJ0-3mix | SI-SNRi: 12.8 -> 13.3 | Helped |
| Train on WSJ0-2&3mix jointly instead of each dataset alone | Conditional TasNet · WSJ0-2mix | SI-SNRi: 15.6 -> 16.3 | Helped |
| Train on WSJ0-3mix jointly with 2mix (vs 3mix alone) | Conditional TasNet · WSJ0-3mix | SI-SNRi: 13.3 -> 13.4 | Helped |
| Train on WSJ0 2-5mix instead of 2&3mix | Conditional TasNet · WSJ0-2mix | SI-SNRi: 16.3 -> 16.7 | Helped |
| Train on WSJ0 2-5mix instead of 2&3mix | Conditional TasNet · WSJ0-3mix | SI-SNRi: 13.4 -> 14.2 | Helped |
| Conditional TasNet 2-5mix vs ConvTasNet on 5 speakers | Conditional TasNet · WSJ0-5mix | SI-SNRi: 6.8 -> 11.7 | Helped |
| Conditional chain Transformer CTC instead of PIT Transformer CTC | Transformer CTC ASR · WSJ0-2mix | WER: 31.2% -> 24.7% | Helped |
| Train with 1, 2 and 3-speaker data instead of 2-speaker only | Conditional-Chain-Transformer CTC · WSJ0-2mix | WER: 24.7% -> 14.9% | Helped |
| Fine-tune TasNet with waveform and CTC alignment conditions (ablation) | Conditional TasNet · WSJ0-2mix | SI-SNRi: 15.2 dB -> 15.5 dB | Helped |
| Fine-tune TasNet with waveform and CTC alignment conditions (ablation) | Conditional TasNet + ASR · WSJ0-2mix | WER: 25.1% -> 17.2% | Helped |
| Joint training of TasNet and ASR (wave condition only) (ablation) | Conditional TasNet + ASR · WSJ0-2mix | SI-SNRi: 15.2 dB -> 12.0 dB | Hurt |
Closest papers by meaning
- Optimal Completion Distillation for Sequence Learning (ICLR 2019 · similarity 0.87)
- Imputer: Sequence Modelling via Imputation and Dynamic Programming (ICML 2020 · similarity 0.87)
- Latent Sequence Decompositions (ICLR 2017 · similarity 0.86)
- An Online Sequence-to-Sequence Model Using Partial Conditioning (NeurIPS 2016 · similarity 0.86)
- Trainable Decoding of Sets of Sequences for Neural Sequence Models (ICML 2019 · similarity 0.85)
- Separate and Reconstruct: Asymmetric Encoder-Decoder for Speech Separation (NeurIPS 2024 · similarity 0.84)
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.