# U2++ MoE: Scaling 4.7x parameters with minimal impact on RTF

arXiv: https://arxiv.org/abs/2404.16407 · HTML: https://mlsanity.com/paper/2404.16407/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in Multimodal MoE models, one of 12 directions in Mixture-of-experts.
- Builds on 15 of its 22 references that are in the map.
- First posted on arXiv 25 Apr 2024.

## Placed in

- Mixture-of-experts / [Multimodal MoE models](https://mlsanity.com/moe/multimodal-moe-models/)

## Builds on (most cited first, 15 of 15)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [Robust Speech Recognition via Large-Scale Weak Supervision](https://mlsanity.com/paper/2212.04356/) (2022)
- [Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer](https://mlsanity.com/paper/1701.06538/) (2017)
- [Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity](https://mlsanity.com/paper/2101.03961/) (2021)
- [Conformer: Convolution-augmented Transformer for Speech Recognition](https://mlsanity.com/paper/2005.08100/) (2020)
- [GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding](https://mlsanity.com/paper/2006.16668/) (2020)
- [Google USM: Scaling Automatic Speech Recognition Beyond 100 Languages](https://mlsanity.com/paper/2303.01037/) (2023)
- [WeNet: Production oriented Streaming and Non-streaming End-to-End Speech Recognition Toolkit](https://mlsanity.com/paper/2102.01547/) (2021)
- [Beyond Distillation: Task-level Mixture-of-Experts for Efficient Inference](https://mlsanity.com/paper/2110.03742/) (2021)
- [SpeechMoE: Scaling to Large Acoustic Models with Dynamic Routing Mixture of Experts](https://mlsanity.com/paper/2105.03036/) (2021)
- [WeNet 2.0: More Productive End-to-End Speech Recognition Toolkit](https://mlsanity.com/paper/2203.15455/) (2022)
- [Language-Routing Mixture of Experts for Multilingual and Code-Switching Speech Recognition](https://mlsanity.com/paper/2307.05956/) (2023)
- [Mixture-of-Expert Conformer for Streaming Multilingual ASR](https://mlsanity.com/paper/2305.15663/) (2023)
- [3M: Multi-loss, Multi-path and Multi-level Neural Networks for speech recognition](https://mlsanity.com/paper/2204.03178/) (2022)

## Built on by (most cited first, 5 of 5)

- [Two Heads are Better Than One: Test-time Scaling of Multi-agent Collaborative Reasoning](https://mlsanity.com/paper/2504.09772/) (2025)
- [TouchASP: Elastic Automatic Speech Perception that Everyone Can Touch](https://mlsanity.com/paper/2412.15622/) (2024)
- [Dynamic Language Group-Based MoE: Enhancing Code-Switching Speech Recognition with Hierarchical Routing](https://mlsanity.com/paper/2407.18581/) (2024)
- [BLR-MoE: Boosted Language-Routing Mixture of Experts for Domain-Robust Multilingual E2E ASR](https://mlsanity.com/paper/2501.12602/) (2025)
- [UME: Upcycling Mixture-of-Experts for Scalable and Efficient Automatic Speech Recognition](https://mlsanity.com/paper/2412.17507/) (2024)

## Abstract (arXiv)

Scale has opened new frontiers in natural language processing, but at a high cost. In response, by learning to only activate a subset of parameters in training and inference, Mixture-of-Experts (MoE) have been proposed as an energy efficient path to even larger and more capable language models and this shift towards a new generation of foundation models is gaining momentum, particularly within the field of Automatic Speech Recognition (ASR). Recent works that incorporating MoE into ASR models have complex designs such as routing frames via supplementary embedding network, improving multilingual ability for the experts, and utilizing dedicated auxiliary losses for either expert load balancing or specific language handling. We found that delicate designs are not necessary, while an embarrassingly simple substitution of MoE layers for all Feed-Forward Network (FFN) layers is competent for the ASR task. To be more specific, we benchmark our proposed model on a large scale inner-source dataset (160k hours), the results show that we can scale our baseline Conformer (Dense-225M) to its MoE counterparts (MoE-1B) and achieve Dense-1B level Word Error Rate (WER) while maintaining a Dense-225M level Real Time Factor (RTF). Furthermore, by applying Unified 2-pass framework with bidirectional attention decoders (U2++), we achieve the streaming and non-streaming decoding modes in a single MoE based model, which we call U2++ MoE. We hope that our study can facilitate the research on scaling speech foundation models without sacrificing deployment efficiency.