# Speech-to-text reasoning and semantic understanding on VoiceBench (test) Source: [Speech-to-text reasoning and semantic understanding on VoiceBench (test)]() Collection index: [llms.txt]() Data updated: 2026-02-26T05:34:52.778Z Evaluation of model performance on speech-to-text reasoning and semantic understanding tasks using the test split of the VoiceBench dataset, measured across various reasoning, instruction following, and general evaluation metrics. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Speech-to-text reasoning and semantic understanding | | Dataset (including any stated split) | VoiceBench (test) | | Metric | Alpaca Eval | | Unit | SCALAR | | Direction | Higher is better | | Results loaded | 13 of 13 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Alpaca Eval](): Higher is better; unit: SCALAR. - [Communication Eval](): Higher is better; unit: SCALAR. - [Wild Voice Score](): Higher is better; unit: SCALAR. - [SD-QA](): Higher is better; unit: PERCENT. - [BBH](): Higher is better; unit: PERCENT. - [Adversarial Benchmark Acc](): Higher is better; unit: PERCENT. - [Instruction Following Eval](): Higher is better; unit: PERCENT. - [OBQA](): Higher is better; unit: PERCENT. - [MMSU Score](): Higher is better; unit: PERCENT. - [Overall Score](): Higher is better; unit: PERCENT. ## Recorded results | Method | Alpaca Eval | Result date | Report paper | | --- | --- | --- | --- | | Whisper + GPT-4o | 4.8 | Not recorded | 6965d1596343fa78c83232db | | GPT-4o | 4.78 | Not recorded | 6965d1596343fa78c83232db | | Qwen3-Omni-30B | 4.74 | Not recorded | 6965d1596343fa78c83232db | | Qwen2.5 | 4.66 | Not recorded | 6965d1596343fa78c83232db | | Whisper + Qwen2.5 | 4.64 | Not recorded | 6965d1596343fa78c83232db | | Qwen2.5 (TN) | 4.61 | Not recorded | 6965d1596343fa78c83232db | | AZEROS | 4.44 | Not recorded | 6965d1596343fa78c83232db | | GLM-4-Voice | 3.97 | Not recorded | 6965d1596343fa78c83232db | | Qwen2.5-Omni | 3.88 | Not recorded | 6965d1596343fa78c83232db | | Phi-4-multimodal | 3.81 | Not recorded | 6965d1596343fa78c83232db | | DeSTA2.5 | 3.73 | Not recorded | 6965d1596343fa78c83232db | | Qwen2-Audio | 3.42 | Not recorded | 6965d1596343fa78c83232db | | Moshi | 2.01 | Not recorded | 6965d1596343fa78c83232db | ### Whisper + GPT-4o Alpaca Eval: 4.8. Result ID: 6996a92a5698665ec4f307d3. [Reporting paper]() Recorded evaluation configuration: {"System Paradigm":"Cascaded System","ASR Module":"Whisper-large-v3","LLM Backbone":"GPT-4o"}. This record does not assert independent reproduction. ### GPT-4o Alpaca Eval: 4.78. Result ID: 6996a92a5698665ec4f307d5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"System Paradigm":"End-to-end Speech-LLM"}. This record does not assert independent reproduction. ### Qwen3-Omni-30B Alpaca Eval: 4.74. Result ID: 6996a92a5698665ec4f307dc. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"System Paradigm":"End-to-end Speech-LLM"}. This record does not assert independent reproduction. ### Qwen2.5 Alpaca Eval: 4.66. Result ID: 6996a92a5698665ec4f307d1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"System Paradigm":"Text Only Model","Backbone":"Qwen2.5-7B-Instruct"}. This record does not assert independent reproduction. ### Whisper + Qwen2.5 Alpaca Eval: 4.64. Result ID: 6996a92a5698665ec4f307d4. [Reporting paper]() Recorded evaluation configuration: {"System Paradigm":"Cascaded System","ASR Module":"Whisper-large-v3","LLM Backbone":"Qwen2.5-7B-Instruct"}. This record does not assert independent reproduction. ### Qwen2.5 (TN) Alpaca Eval: 4.61. Result ID: 6996a92a5698665ec4f307d2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"System Paradigm":"Text Only Model","Backbone":"Qwen2.5-7B-Instruct"}. This record does not assert independent reproduction. ### AZEROS Alpaca Eval: 4.44. Result ID: 6996a92a5698665ec4f307dd. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"System Paradigm":"End-to-end Speech-LLM"}. This record does not assert independent reproduction. ### GLM-4-Voice Alpaca Eval: 3.97. Result ID: 6996a92a5698665ec4f307d8. [Reporting paper]() Recorded evaluation configuration: {"System Paradigm":"End-to-end Speech-LLM"}. This record does not assert independent reproduction. ### Qwen2.5-Omni Alpaca Eval: 3.88. Result ID: 6996a92a5698665ec4f307db. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"System Paradigm":"End-to-end Speech-LLM"}. This record does not assert independent reproduction. ### Phi-4-multimodal Alpaca Eval: 3.81. Result ID: 6996a92a5698665ec4f307d7. [Reporting paper]() Recorded evaluation configuration: {"System Paradigm":"End-to-end Speech-LLM"}. This record does not assert independent reproduction. ### DeSTA2.5 Alpaca Eval: 3.73. Result ID: 6996a92a5698665ec4f307da. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"System Paradigm":"End-to-end Speech-LLM"}. This record does not assert independent reproduction. ### Qwen2-Audio Alpaca Eval: 3.42. Result ID: 6996a92a5698665ec4f307d9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"System Paradigm":"End-to-end Speech-LLM"}. This record does not assert independent reproduction. ### Moshi Alpaca Eval: 2.01. Result ID: 6996a92a5698665ec4f307d6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"System Paradigm":"End-to-end Speech-LLM"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.