MAGICDATA-DIALECT-SICHUANESE-NAVIGATION: Sichuan–Chongqing Dialect Navigation Speech Dataset

Magic Data has released the MAGICDATA-DIALECT-SICHUANESE-NAVIGATION Sichuan–Chongqing dialect navigation speech recognition dataset on the MagicHub.com, for developers, researchers, and speech technology practitioners.

Dataset Introduction

The dataset contains 32.5 hours of audio and targets Chinese dialect automatic speech recognition (ASR) in navigation, ride-hailing, and travel-inquiry scenarios. Centered on Sichuan–Chongqing accent/dialect speech, it comprises speech content designed around real travel intents, covering common interactions such as route origin and destination descriptions, vehicle-type selection, ride booking, platform promotions, and order confirmation.

The dataset is suitable for evaluating the transcription accuracy, domain-keyword recognition, and accent robustness of speech recognition systems on Sichuan-dialect navigation speech. Compared with general read-speech corpora, navigation-domain text contains a large number of place names, travel-service terms, and natural spoken expressions, and therefore reflects more directly the practical recognition challenges faced by in-car voice assistants, map navigation, ride-hailing platforms, and local-life services.

Dataset Overview

• Data type: Automatic speech recognition (ASR)
• Language: Chinese dialect/accent
• Total duration: 32.5 hours
• Total size: 2.84 GB
• Coverage: 15 cities in the Sichuan–Chongqing region
• Recording device: Mobile phone
• Recording environment: Quiet indoor

![[Codex 图像 2026年8月31日 15_36_55.svg|156]]![[city-distribution.png|234]]

Figure 1: Gender distribution of dataset speakers ; Figure 2: City distribution of dataset speakers

Data Composition

Dataset name	Primary use	Audio duration	File size	Coverage
MAGICDATA-DIALECT-SICHUANESE-NAVIGATION-TRAININGSET	Model development, parameter tuning, validation/evaluation, and error analysis for speech recognition on Sichuan-dialect navigation speech	13 hours	1.14 GB	12 cities in the Sichuan–Chongqing region: Chengdu, Chongqing, Leshan, Yibin, Luzhou, Zigong, Neijiang, Ziyang, Mianyang, Nanchong, Dazhou, Guang'an
MAGICDATA-DIALECT-SICHUANESE-NAVIGATION-TESTSET-PART1	Model development, parameter tuning, validation/evaluation, and error analysis for speech recognition on Sichuan-dialect navigation speech	13 hours	1.12 GB	12 cities in the Sichuan–Chongqing region: Chengdu, Chongqing, Leshan, Yibin, Luzhou, Zigong, Neijiang, Ziyang, Mianyang, Nanchong, Dazhou, Guang'an
MAGICDATA-DIALECT-SICHUANESE-NAVIGATION-TESTSET-PART2	Model development, parameter tuning, validation/evaluation, and error analysis for speech recognition on Sichuan-dialect navigation speech	6.5 hours	579 MB	6 cities in Sichuan: Chengdu, Zigong, Neijiang, Bazhong, Guangyuan, Suining

Applications

• Testing and benchmark evaluation of Chinese ASR models for Sichuan dialect/accent.
• Scenario-based recognition evaluation of navigation, in-car voice assistant, and ride-hailing voice interaction systems.
• Analysis of recognition performance on navigation-domain keywords such as place names, vehicle types, times, and promotions.
• Validation of the voice interaction experience for Sichuan-localized intelligent mobility services.

Audio and Annotation Format

Item	Specification
Audio Parameters	16 kHz,16 bits,Mono
File Forma	WAV (PCM),TXT (UTF-8)
Speech type	Read speech

Download

The dataset is available from the MagicHub open-source community:

MAGICDATA-DIALECT-SICHUANESE-NAVIGATION-TESTSET-PART1: https://magichub.com/datasets/magicdata-dialect-sichuanese-navigation-testset-part1/ MAGICDATA-DIALECT-SICHUANESE-NAVIGATION-TESTSET-PART2: https://magichub.com/datasets/magicdata-dialect-sichuanese-navigation-testset-part2/ MAGICDATA-DIALECT-SICHUANESE-NAVIGATION-TRAININGSET: https://magichub.com/datasets/magicdata-dialect-sichuanese-navigation-trainingset/

Usage

The dataset is suitable for evaluating the transcription accuracy, domain-keyword recognition, and accent robustness of speech recognition systems on Sichuan-dialect navigation speech. Compared with general read-speech corpora, navigation-domain text contains a large number of place names, travel-service terms, and natural spoken expressions, and therefore reflects more directly the practical recognition challenges faced by in-car voice assistants, map navigation, ride-hailing platforms, and local-life services.

Before using the dataset, please read the license and data usage notes in the repository, and comply with applicable laws, regulations, and privacy protection requirements.

License: MAGIC DATA OPEN-SOURCE LICENSE

Contact Us

If you have any questions or suggestions, please feel free to submit an Issue in the GitHub repository or contact us by email.