Latest

Artificial Intelligence

This is VibeVoice, Microsoft's bet on open source voice AI

VibeVoice is Microsoft's bet on open source voice AI, with models capable of generating natural, multi-speaker audio

This is VibeVoice, Microsoft's bet on open source voice AI
This is VibeVoice, Microsoft's bet on open source voice AI

VibeVoice, developed by Microsoft and released as open source, has become one of the most prominent projects in the field of audio and AI. The system has gained rapid popularity on GitHub, where it already has more than 29,000 stars. The initiative brings together text-to-speech (TTS) and automatic speech recognition (ASR) models in a single ecosystem, with the aim of making it easier to build voice-based applications. Its growing adoption reflects the industry's interest in more accessible and powerful tools in this field.

One of the most striking aspects of VibeVoice is its ability to generate high-quality conversational audio over long periods of time. Its main model can produce up to 90 minutes of continuous audio with up to four different speakers, maintaining consistency between voices, natural turn-taking and emotional expressiveness. This opens up new possibilities for creating podcasts, audiobooks and complex narrative content.

How it works

At a technical level, VibeVoice combines a language model that interprets the context and the evolution of the dialogue, while other models convert that information into realistic audio, refining it progressively until it achieves a quality close to that of a human voice.

Trump's six priorities for regulating AI in the United States

The family includes different versions adapted to different uses. The main model focuses on long, multi-speaker audio generation, while VibeVoice-Realtime is optimized for real-time applications with low latency. VibeVoice-ASR, for its part, can transcribe up to 60 minutes of audio in a single pass, identify the speakers and add timestamps.

In recent weeks, the open-source community has accelerated its adoption, notably "Vibing", a voice input system for macOS and Windows based on this technology. The project is distributed under the MIT license, which allows its free use, modification and integration. Microsoft has also made adoption easier by integrating it into Hugging Face's Transformers library, one of the most widely used standards in artificial intelligence.

By betting on open source, Microsoft aims to attract developers and accelerate the creation of new applications, while at the same time driving use of its cloud infrastructure.

Maria Gonzalez

Author

Maria Gonzalez

Journalist specializing in cryptocurrencies, artificial intelligence and emerging technology, with experience in trend analysis, innovation coverage and writing features that connect the digital economy with the real world.

The latest
From Huang to Botín, from Zara to Byron: corporate responsibility in global capitalism

Keep reading