REVIEW 8 cited by
Zipformer: A faster and better encoder for automatic speech recognition
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
The Conformer has become the most popular encoder model for automatic speech recognition (ASR). It adds convolution modules to a transformer to learn both local and global dependencies. In this work we describe a faster, more memory-efficient, and better-performing transformer, called Zipformer. Modeling changes include: 1) a U-Net-like encoder structure where middle stacks operate at lower frame rates; 2) reorganized block structure with more modules, within which we re-use attention weights for efficiency; 3) a modified form of LayerNorm called BiasNorm allows us to retain some length information; 4) new activation functions SwooshR and SwooshL work better than Swish. We also propose a new optimizer, called ScaledAdam, which scales the update by each tensor's current scale to keep the relative change about the same, and also explictly learns the parameter scale. It achieves faster convergence and better performance than Adam. Extensive experiments on LibriSpeech, Aishell-1, and WenetSpeech datasets demonstrate the effectiveness of our proposed Zipformer over other state-of-the-art ASR models. Our code is publicly available at https://github.com/k2-fsa/icefall.
Forward citations
Cited by 8 Pith papers
-
UniVoice: Unifying Autoregressive ASR and Flow-Matching based TTS with Large Language Models
A single LLM can do ASR and zero-shot TTS on continuous speech features by switching between causal and bidirectional attention, reaching competitive but not state-of-the-art results.
-
XEmoRAG: Cross-Lingual Emotion Transfer with Controllable Intensity Using Retrieval-Augmented Generation
XEmoRAG synthesizes Thai speech with emotions cloned from Chinese reference audio by retrieving matching Thai prompts and aligning prosody with flow matching, outperforming baseline TTS in emotion similarity and intel...
-
Raon-Speech Technical Report
A 9B SpeechLM trained on 1.38M hours of English/Korean data plus a full-duplex chat extension trained on 119K hours of time-aligned dialogue outperform same-size audio models on speech tasks and FDB turn-taking metrics.
-
SketchSong: Hierarchical Song Generation with Sketch Planning and Fine-Grained Multi-Track Modeling
SketchSong uses temporal sketch planning with high-level tokens and explicit modeling of four tracks (vocals, bass, drums, other) to generate more coherent songs than baselines.
-
Exploring Cross-Utterance Speech Contexts for Conformer-Transducer Speech Recognition Systems
Adding cross-utterance audio context to Conformer-Transducer ASR reduces WER/CER by 0.5 to 1.1 absolute points on four benchmarks, and a splicing-based batch scheme cuts training time by up to about 19%.
-
Towards Hallucination-Free Music: A Reinforcement Learning Preference Optimization Framework for Reliable Song Generation
PER-based preference optimization (DPO, PPO, GRPO) reduces lyric-to-song hallucination in an audio language model, with the largest gains from DPO plus reject sampling.
-
BoSS: Beyond-Semantic Speech
Current spoken-language models perform poorly on a new five-task evaluation of beyond-semantic speech signals, including dialect, emotion, age, and non-verbal cues.
-
From Objectives to Applications: Aligning Architectural Biases in Audio Self-Supervised Learning
A survey that organizes audio SSL into five objective paradigms, relates their demands to architectural biases, and interprets downstream applications as tests of generalization.
Discussion (0). Sign in to comment.