Dynamic Insight Beating AI News: Ant Ling opensourced the Ming-3.0 speech series, covering speech synthesis, audio understanding, and speech recognition, with a total of 5 models—2 TTS, 2 Audio, and 1 ASR—all under the MIT license. Developers can combine them to achieve voice input, content analysis, and voice response.

The standout is Ming-3.0-tts-0.5B, with only 500 million parameters. Users can instruct the AI via natural language to speak with suppressed anger, emphasize specific words, or adjust dialogue rhythm. It supports voice cloning, text-designed voices, and inserting 14 types of sound tags such as light laughter or sighs at designated positions. In Artificial Analysis's same-voice blind listening test, this model ranked first among open-source models with a 1069 Elo score.

Ming-3.0-audio-flash, responsible for audio understanding, is significantly larger, with a total of 126 billion parameters and approximately 6.5 billion activated per use. It can directly process meeting recordings, answer questions about discussion topics, and summarize agreements, disagreements, and action items without requiring a separate transcript.

The other model, Ming-3.0-asr-tiny, specializes in speech recognition, supporting names, locations, industry terms, and prior conversation context to assist in recognizing dialects and rare words. In four categories of speech recognition tests published by Ant, its average error rate was 4.41%.

Dynamic Insight Beating AI News: Wang Ziru, founder of ZEALER and former Gree executive, has unveiled his team's first AI product, VIVA, focused on personal health management.

It integrates three virtual roles—nutritionist, fitness coach, and life assistant—into a single chat interface. Users simply take a photo of their meal or verbally state what they've eaten, and the AI estimates calorie intake and logs nutritional data.

The fitness coach can schedule workouts based on the user's daily routine, guide exercises, and track progress. The life assistant connects with Apple Health, using sleep and activity data to adjust daily recommendations—for example, reducing workout intensity if the user had insufficient sleep the previous night. These three roles share information within the conversation, eliminating the need to switch between different tools.

This product has an interesting backstory. In June this year, Wang Ziru demonstrated an early prototype of his AI health management system. He revealed that he personally used AI-assisted methods to lose weight, dropping from 86 kg to 74 kg. During development, he aimed to solve two key challenges: the complexity of Chinese cooking, making calorie estimation difficult; and fragmented data from smart wearables, which required manual consolidation before AI could analyze it.

In August, Wang Ziru mentioned that the startup software would be launched no later than September. Now, the team plans to offer live experiences of VIVA at Xiaohongshu's "Mañana Tomorrow Park" in Haikou from October 23 to 25.