Open Source Audio Models Overview (February 2026)

The landscape of audio models is rapidly evolving, with frequent new releases, including Qwen3 TTS. This article aims to provide an overview of the best open-source audio models currently available.

The goal is to gather user experiences with different ASR (Automatic Speech Recognition), TTS (Text-to-Speech), STT (Speech-to-Text), and text-to-music models, inviting them to share their setups, usage contexts (personal or professional), tools, and frameworks used.

Given the subjectivity in evaluating these models, users are encouraged to provide detailed descriptions of their setup and usage. Closed models, such as Elevenlabs v3, appear to maintain an advantage in terms of performance, especially for production uses that require stability and management of long audio sequences. Therefore, empirical comparisons are particularly useful.

Rules:

  • Models must have open weights.