File size: 6,918 Bytes
40bdebd be53961 40bdebd 39c2089 7c5e63b 39c2089 7c5e63b 39c2089 7c5e63b 39c2089 7c5e63b c9dde62 7c5e63b c9dde62 7c5e63b c9dde62 be53961 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 |
---
license: apache-2.0
pipeline_tag: automatic-speech-recognition
library_name: transformers
---
# 1. Step-Audio-Chat
This repository contains the Multimodal Large Language Model (LLM) component of Step-Audio. It is a 130 billion parameter multimodal LLM that is responsible for understanding and generating human speech. The model is specifically designed to seamlessly integrate functions such as speech recognition, semantic understanding, dialogue management, voice cloning, and speech generation.
## 2. Evaluation
### 2.1 LLM judge metrics(GPT-4o) on [**StepEval-Audio-360**](https://huggingface.co/datasets/stepfun-ai/StepEval-Audio-360)
<table>
<caption>Comparison of fundamental capabilities of voice chat on the StepEval-Audio-360.</caption>
<thead>
<tr>
<th>Model</th>
<th style="text-align:center">Factuality (% ↑)</th>
<th style="text-align:center">Relevance (% ↑)</th>
<th style="text-align:center">Chat Score ↑</th>
</tr>
</thead>
<tbody>
<tr>
<td>GLM4-Voice</td>
<td style="text-align:center">54.7</td>
<td style="text-align:center">66.4</td>
<td style="text-align:center">3.49</td>
</tr>
<tr>
<td>Qwen2-Audio</td>
<td style="text-align:center">22.6</td>
<td style="text-align:center">26.3</td>
<td style="text-align:center">2.27</td>
</tr>
<tr>
<td>Moshi<sup>*</sup></td>
<td style="text-align:center">1.0</td>
<td style="text-align:center">0</td>
<td style="text-align:center">1.49</td>
</tr>
<tr>
<td><strong>Step-Audio-Chat</strong></td>
<td style="text-align:center"><strong>66.4</strong></td>
<td style="text-align:center"><strong>75.2</strong></td>
<td style="text-align:center"><strong>4.11</strong></td>
</tr>
</tbody>
</table>
*Note: Moshi are marked with "\*" and should be considered for reference only.
### 2.2 Public Test Set
<table>
<thead>
<tr>
<th>Model</th>
<th style="text-align:center">Llama Question</th>
<th style="text-align:center">Web Questions</th>
<th style="text-align:center">TriviaQA*</th>
<th style="text-align:center">ComplexBench</th>
<th style="text-align:center">HSK-6</th>
</tr>
</thead>
<tbody>
<tr>
<td>GLM4-Voice</td>
<td style="text-align:center">64.7</td>
<td style="text-align:center">32.2</td>
<td style="text-align:center">39.1</td>
<td style="text-align:center">66.0</td>
<td style="text-align:center">74.0</td>
</tr>
<tr>
<td>Moshi</td>
<td style="text-align:center">62.3</td>
<td style="text-align:center">26.6</td>
<td style="text-align:center">22.8</td>
<td style="text-align:center">-</td>
<td style="text-align:center">-</td>
</tr>
<tr>
<td>Freeze-Omni</td>
<td style="text-align:center">72.0</td>
<td style="text-align:center">44.7</td>
<td style="text-align:center">53.9</td>
<td style="text-align:center">-</td>
<td style="text-align:center">-</td>
</tr>
<tr>
<td>LUCY</td>
<td style="text-align:center">59.7</td>
<td style="text-align:center">29.3</td>
<td style="text-align:center">27.0</td>
<td style="text-align:center">-</td>
<td style="text-align:center">-</td>
</tr>
<tr>
<td>MinMo</td>
<td style="text-align:center">78.9</td>
<td style="text-align:center">55.0</td>
<td style="text-align:center">48.3</td>
<td style="text-align:center">-</td>
<td style="text-align:center">-</td>
</tr>
<tr>
<td>Qwen2-Audio</td>
<td style="text-align:center">52.0</td>
<td style="text-align:center">27.0</td>
<td style="text-align:center">37.3</td>
<td style="text-align:center">54.0</td>
<td style="text-align:center">-</td>
</tr>
<tr>
<td><strong>Step-Audio-Chat</strong></td>
<td style="text-align:center"><strong><i>81.0</i></strong></td>
<td style="text-align:center"><strong>75.1</strong></td>
<td style="text-align:center"><strong>58.0</strong></td>
<td style="text-align:center"><strong>74.0</strong></td>
<td style="text-align:center"><strong>86.0</strong></td>
</tr>
</tbody>
</table>
*Note: Results marked with "\*" on TriviaQA dataset are considered for reference only.*
*TriviaQA dataset marked with "\*" indicates results are for reference only.*
### 2.3 Audio instruction following
<table>
<thead>
<tr>
<th rowspan="2">Category</th>
<th colspan="2" style="text-align:center">Instruction Following</th>
<th colspan="2" style="text-align:center">Audio Quality</th>
</tr>
<tr>
<th style="text-align:center">GLM-4-Voice</th>
<th style="text-align:center">Step-Audio</th>
<th style="text-align:center">GLM-4-Voice</th>
<th style="text-align:center">Step-Audio</th>
</tr>
</thead>
<tbody>
<tr>
<td>Languages</td>
<td style="text-align:center">1.9</td>
<td style="text-align:center">3.8</td>
<td style="text-align:center">2.9</td>
<td style="text-align:center">3.3</td>
</tr>
<tr>
<td>Role-playing</td>
<td style="text-align:center">3.8</td>
<td style="text-align:center">4.2</td>
<td style="text-align:center">3.2</td>
<td style="text-align:center">3.6</td>
</tr>
<tr>
<td>Singing / RAP</td>
<td style="text-align:center">2.1</td>
<td style="text-align:center">2.4</td>
<td style="text-align:center">2.4</td>
<td style="text-align:center">4</td>
</tr>
<tr>
<td>Voice Control</td>
<td style="text-align:center">3.6</td>
<td style="text-align:center">4.4</td>
<td style="text-align:center">3.3</td>
<td style="text-align:center">4.1</td>
</tr>
</tbody>
</table>
## 3. More information
For more information, please refer to our repository: [Step-Audio](https://github.com/stepfun-ai/Step-Audio) and the paper [Step-Audio: Unified Understanding and Generation in Intelligent Speech Interaction](https://hf.co/papers/2502.11946). |