Back to AI information
Alibaba releases Tongyi Fun: an enterprise-grade audio foundation model that combines Fun-ASR and Fun-CosyVoice

Alibaba releases Tongyi Fun: an enterprise-grade audio foundation model that combines Fun-ASR and Fun-CosyVoice

AI information Admin 92 views

Alibaba Tongyi has launched Tongyi Fun, an enterprise-grade audio foundation model. Comprising Fun-ASR and Fun-CosyVoice, it aims to achieve full-chain capabilities from "understanding to transcription to speaking" in real-world enterprise environments. Official information indicates that Fun-ASR is trained with tens of millions of hours of real speech, covering industry terminology and noisy scenarios in finance, technology, manufacturing, and other industries. Fun-CosyVoice provides multilingual and multi-timbre synthesis and cross-lingual voice cloning, emphasizing stable prosody and natural expression.

In terms of engineering implementation, Tongyi Fun employs contextual enhancement and RAG retrieval to integrate enterprise knowledge and terminology into the recognition and translation process, reducing cross-language interference and artifacts. To meet practical needs, the product supports meeting minutes, customer service quality control, voice bots, and multimedia content production. It also provides an extensible API and customizable training paths for easy integration with existing systems.

Frequently Asked Questions

Q: What are the core components of Tongyi Fun?

A: It consists of Fun-ASR (speech recognition) and Fun-CosyVoice (speech synthesis), which are responsible for understanding and generation respectively.

Q: Why introduce RAG?

A: By searching the enterprise knowledge base and glossary, we can enhance the context, make transcription and translation more accurate, and reduce illusions and misjudgments.

Q: What is the coverage of languages and tones?

A: It supports multiple languages, multiple dialects, and a variety of anthropomorphic voices, and provides cross-language voice cloning to maintain speaker consistency.

Q: What are the applicable enterprise scenarios?

A: Meeting transcription and summarization, customer service quality inspection and voice robots, training and content production, compliance review, etc.

Q: Is open source or access path provided?

A: Components such as CosyVoice have open source implementations; enterprise-level access and consoles are provided on the Alibaba Cloud Tongyi product page.

TongyiFun Enterprise Audio TongyiFun full-link listening and speaking TongyiFunFunASR recognition TongyiFunFunCosyVoice Synthesis TongyiFun multilingual and multi-dialect TongyiFun Multi-Timbre Cloning TongyiFun cross-language cloning TongyiFun context enhancement TongyiFunRAG Search TongyiFun industry terminology adaptation TongyiFun is robust in noisy scenes TongyiFun meeting minutes TongyiFun Customer Service Quality Inspection TongyiFun Voice Robot TongyiFun content production TongyiFun real-time transcription TongyiFun low-latency conversation TongyiFun Translation Enhancement TongyiFun terminology injection TongyiFun knowledge base alignment TongyiFun Hallucination Suppression TongyiFun stable rhythm TongyiFun Natural Expression TongyiFunAPI access TongyiFun customizable training TongyiFun Enterprise Integration TongyiFun Data Compliance TongyiFun speaker keeps TongyiFun long audio TongyiFun multi-modal linkage TongyiFun Financial Scenario TongyiFun manufacturing scene TongyiFun Technology Scene TongyiFun quality inspection rules synchronization TongyiFun Summary Generation TongyiFun keyword extraction TongyiFun Acoustic Robustness TongyiFun end-to-end process TongyiFun Model Governance TongyiFun deployment flexibility TongyiFun's costs are controllable TongyiFun console management TongyiFun Live Streaming TongyiFun Batch Transcription TongyiFun Compliance Audit TongyiFun Multi-channel Phone TongyiFun Conference Assistant TongyiFun has trained corpus for tens of millions of hours TongyiFunSDK toolchain TongyiFun Ecosystem Compatibility

Recommended Tools

More