# Alibaba Tongyi Launches End-to-End Speech Recognition Model Fun-ASR 1.5

- Category: Flash (快讯)
- Published: 2026-04-20T11:56:00.000Z
- Source date: 2026-04-20
- HTML: https://www.88z.ai/en/news/flash-20260420-f9b1968cfcc76431
- Markdown: https://www.88z.ai/en/news/flash-20260420-f9b1968cfcc76431.md
- Other language: https://www.88z.ai/news/flash-20260420-f9b1968cfcc76431.md

Alibaba Tongyi has launched Fun-ASR 1.5, an end-to-end speech recognition model supporting 30 languages and seven major Chinese dialects. It features significantly reduced character error rates and specialized ancient poetry recognition. Built on MoE architecture with smart punctuation and text normalization, it is now available on Alibaba Cloud Bailian and ModelScope.

## At a glance

- Supports 30 languages and seven Chinese dialects, with a 56.2% reduction in character error rate
- Uses MoE architecture with 97% accuracy in ancient poetry recognition, available on Alibaba Cloud Bailian and ModelScope

## The update

Alibaba Tongyi has officially released Fun-ASR 1.5, an end-to-end speech recognition large model. This single model covers thirty languages, supporting automatic language switching and mixed-language speech scenarios. For Chinese processing, it comprehensively covers seven major dialect systems, reducing the character error rate by 56.2% compared to the previous version. Additionally, the new release enhances specialized ancient poetry recognition, achieving an accuracy rate of 97%. In terms of technical architecture, Fun-ASR 1.5 adopts a Mixture of Experts (MoE) structure. It supports intelligent punctuation prediction and performs text normalization for content such as numbers and dates. The model is now officially live on Alibaba Cloud Bailian and ModelScope for developer use.

## Sources

- [千问大模型](https://mp.weixin.qq.com/s/xPEstUiJ4IC3BOFVsdvRLA) (2026-04-20)

This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
