# xid32/SoundMind

We introduce the Audio Logical Reasoning (ALR) dataset, consisting of 6,446 text-audio annotated samples specifically designed for complex reasoning tasks. Building on this resource, we propose SoundMind, a rule-based reinforcement learning (RL) algorithm tailored to endow audio language models (ALMs) with deep bimodal reasoning abilities.

Repository: https://github.com/xid32/SoundMind
Canonical: https://ross.abutalabs.com/products/soundmind
Homepage: https://arxiv.org/abs/2506.12935
Language: Python
License: MIT
License Family: permissive
Topics: audio-language-model, audio-reasoning, dataset, reinforcement-learning
Last push: 2025-11-26T22:19:44+00:00

## Health v2 (maintenance only)
Score: 43/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 54, release rhythm 35, longevity 31
- inputs: {"age_days": 446, "days_push": 280, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1113, forks 131 (observed 2026-08-28T04:03:38.068189+00:00)

## What it is
SoundMind is an Audio Logical Reasoning (ALR) dataset of 6,446 audio-text annotated samples with chain-of-thought reasoning, paired with a rule-based reinforcement learning training framework for audio-language models built on verl. It was used to fine-tune Qwen2.5-Omni-7B, achieving strong improvements on audio reasoning benchmarks.

## Use cases
- train audio language models for logical reasoning with reinforcement learning
- download an annotated audio-text reasoning dataset with chain-of-thought
- benchmark audio-language models on complex reasoning tasks
- fine-tune Qwen2.5-Omni with rule-based RL
- research bimodal audio-text reasoning in LLMs

## When to choose
- you need reasoning-oriented audio training data with chain-of-thought annotations
- you want to apply rule-based RL to audio-language models using verl
- you are evaluating audio models on logical reasoning benchmarks

## When to avoid
- you lack multi-GPU H100/H800-class hardware for training
- you need general audio transcription or speech recognition rather than reasoning
- you want a production-ready inference service rather than a research codebase

## Facets
- artifact type: dataset
- maturity: active
- function: machine-learning, llm-training, audio-processing, speech-recognition
- domain: machine-learning, artificial-intelligence
- platform: python
- tags: audio-language-model, reinforcement-learning, chain-of-thought, benchmark, qwen2.5-omni, verl, emnlp-2025, audio, natural-language-processing, gpu, linux

## Member repositories
- xid32/SoundMind (main) score 43

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:38.068189+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:42:38.275990+00:00, confidence not recorded.
  - readme: https://github.com/xid32/SoundMind (fetched 2026-08-28T04:03:38.068189+00:00, sha 4591bc772939)
  - homepage: https://arxiv.org/abs/2506.12935 (fetched 2026-08-29T12:46:41.767767+00:00, sha 10a06cc3b52b)
  - site_page: https://info.arxiv.org/about/donate.html (fetched 2026-08-29T12:46:41.795740+00:00, sha cca9c3a11c56)
  - site_page: https://info.arxiv.org/about/ourmembers.html (fetched 2026-08-29T12:46:41.799386+00:00, sha 47cbc55ff1de)
  - site_page: https://info.arxiv.org/about (fetched 2026-08-29T12:46:41.801229+00:00, sha a1f16f915a9a)
  - site_page: https://info.arxiv.org/labs/index.html (fetched 2026-08-29T12:46:41.797669+00:00, sha b14a8d05a0ec)
- Data as of 2026-08-30T08:39:29.467469+00:00.