Research map / Architectures and efficiency / RNNs and normalization
Understanding Adaptive, Multiscale Temporal Integration In Deep Speech Recognition Systems
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 |
|---|---|---|---|
| train the network versus random initialization (early layers Conv1-LSTM2) (ablation) | DeepSpeech2 · integration windows | integration window: Conv1 median 58 ms vs random LSTM medians 167-204 ms | Helped |
| train the network versus random initialization (later layers LSTM3-softmax) (ablation) | DeepSpeech2 · integration windows | integration window: LSTM5 median 592 ms | Mixed |
| stretch/compress speech by 20% in a trained network, early versus late layers | DeepSpeech2 · adaptation index | adaptation index: 0.19/0.21 (LSTM1) -> 0.84/0.63 (LSTM3) | Mixed |
| time-stretch/compress stimuli in randomly initialized model (ablation) | DeepSpeech2 (random init) · adaptation index | adaptation index | No clear effect |
| stretch speech instead of compress it | DeepSpeech2 · LibriSpeech test-clean | word error rate (%): 7.6 -> 8.3 (stretched), 10.4 (compressed) | Hurt |
| use fully convolutional architecture (QuartzNet-based) instead of recurrent (ablation) | deep speech recognition models · adaptation index | adaptation index | Hurt |
Closest papers by meaning
- Large language models transition from integrating across position-yoked, exponential windows to structure-yoked, power-law windows (NeurIPS 2023 · similarity 0.91)
- Revealing Vision-Language Integration in the Brain with Multimodal Networks (ICML 2024 · similarity 0.85)
- Prospective Coding Improves Learning in Deep Continuous-Time Recurrent Networks (NeurIPS 2026 · similarity 0.85)
- OWLS: Scaling Laws for Multilingual Speech Recognition and Translation Models (ICML 2025 · similarity 0.84)
- Dynamical modeling of nonlinear latent factors in multiscale neural activity with real-time inference (NeurIPS 2025 · similarity 0.84)
- RTFS-Net: Recurrent Time-Frequency Modelling for Efficient Audio-Visual Speech Separation (ICLR 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.