AISHELL-2: Transforming Mandarin ASR Research Into Industrial Scale

By Jiayu Du, Xingyu Na, Xuechen Liu, Hui Bu. 2018

368 citations recorded by the source corpus.

AISHELL-1 is by far the largest open-source speech corpus available for Mandarin speech recognition research. It was released with a baseline system containing solid training and testing pipelines for Mandarin ASR. In AISHELL-2, 1000 hours of clean read-speech data from iOS is published, which is free for academic usage. On top of AISHELL-2 corpus, an improved recipe is developed and released, containing key components for industrial applications, such as Chinese word segmentation, flexible vocabulary expension and phone set transformation etc. Pipelines support various state-of-the-art techniques, such as time-delayed neural networks and Lattic-Free MMI objective funciton. In addition, we also release dev and test data from other channels(Android and Mic). For research community, we hope that AISHELL-2 corpus can be a solid resource for topics like transfer learning and robust ASR. For industry, we hope AISHELL-2 recipe can be a helpful reference for building meaningful industrial systems and products.

Subfields: Artificial Intelligence, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Shiliang Zhang
2. Zhuo Chen
3. Jian Wu
4. Lei Xie
5. Zhihao Du
6. Kai Hu
7. Siqi Zheng
8. Ziyang Ma

In Signal Processing

1. Jian Wu
2. Lei Xie
3. Zhuo Chen
4. Shiliang Zhang
5. Zhihao Du
6. Kai Hu
7. Siqi Zheng
8. Jun Du

Citations in the corpus

• FunCodec: A Fundamental, Reproducible and Integrable Open-Source Toolkit for Neural Speech Codec (2023, methodology): “To assess the generalization ability of our generalized model, we evaluate it on the test sets of multiple open-source corpora, including Librispeech [28], aishell-1 [29], aishell-2 [30], Wenet [31] and Gigaspeech [32].”
• A Survey on Neural Speech Synthesis (2021, neutral): “1 English CSMSC [15] 12 1 48 Mandarin HKUST [205] 200 2100 8 Mandarin AISHELL-1 [28] 170 400 16 Mandarin AISHELL-2 [69] 1000 1991 44.”
• WENETSPEECH: A 10000+ Hours Multi-Domain Mandarin Corpus for Speech Recognition (2021, methodology): “Comparing with English corpora, the largest open source Mandarin speech data is AIShell-2 [23], including 1000 hours speech recorded in a quiet environment and having a state-of-the-art character error rate result of 5.35%.”
• Qwen-Audio: Advancing Universal Audio Understanding via Unified Large-Scale Audio-Language Models (2023, methodology): “(Du et al., 2018) test Librispeech (Panayotov et al., 2015) In order to facilitate versatile input from both audio and pure text modalities within multi-turn dialogues, we use a combination of audio-centric instruction data mentioned above and pure text instruction data during this training process.”
• AISHELL-4: An Open Source Dataset for Speech Enhancement, Separation, Recognition and Speaker Diarization in Conference Scenario (2021, methodology): “For Mandarin speech corpus, the commonly used open source datasets for speech recognition include AISHELL1 [20], ASHELL-2 [21], aidatatang(2), etc.”
• LauraGPT: Listen, Attend, Understand, and Regenerate Audio with GPT (2023, methodology): “Due to the limited data volumes of BSTC and CoVoST 2, we further augment the training set by translating AISHELL-1 and AISHELL-2 datasets into English and translating LibriSpeech dataset into Chinese using a publicly available text translation model (Wei et al., 2022b).”