Safetensors
Jayjay55 commited on
Commit
909629d
·
verified ·
1 Parent(s): b8c8adc

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +460 -0
README.md CHANGED
@@ -1,3 +1,463 @@
1
  ---
2
  license: apache-2.0
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: apache-2.0
3
  ---
4
+
5
+ <h1>FireRedAudio</h1>
6
+ <div align="center">
7
+ <p>
8
+ </p>
9
+ <p>
10
+ Official PyTorch code for <br>
11
+ <b><em>FireRedAudio: A General-Purpose Audio Language Model with Decoupled Continuous Representations for Understanding and Generation</em></b>
12
+ </p>
13
+ <p>
14
+ </p>
15
+ <a href="#"><img src="https://img.shields.io/badge/Paper-ArXiv-red" alt="technical report"></a>
16
+ <a href="https://fireredteam.github.io/demos/fireredaudio/"><img src="https://img.shields.io/badge/Demo-Page-lightgrey" alt="version"></a>
17
+ <a href="https://huggingface.co/FireRedTeam/FireRedAudio"><img src="https://img.shields.io/badge/Hugging%20Face-Model%20Page-yellow" alt="HF-model"></a>
18
+ <a href="LICENSE"><img src="https://img.shields.io/badge/License-Apache%202.0-blue.svg" alt="Apache-2.0"></a>
19
+ </div>
20
+
21
+
22
+ ## Overview
23
+
24
+ > **One model to listen, understand, reason, speak, and edit.**
25
+
26
+ **FireRedAudio** is a general-purpose audio language model built on a shared **9B-parameter LLM** with **decoupled continuous representations**: an Audio Encoder handles understanding, while a RedAE pathway handles generation. A single model supports **ASR, audio understanding, zero-shot TTS, instruct TTS, semantic/acoustic speech editing, and accurate temporal grounding over recordings up to one hour long**.
27
+
28
+ <div align="center">
29
+ <img src="https://raw.githubusercontent.com/FireRedTeam/FireRedAudio/main/assets/fireredaudio_logo.png" alt="FireRedAudio Logo" width="90%">
30
+ <br>
31
+ </div>
32
+
33
+
34
+ ## Highlights ✨
35
+
36
+ * 🧩 **Purpose-built representations, one shared backbone** — The Audio Encoder pathway serves understanding, while the RedAE-Patch pathway serves speech generation. Their representations remain decoupled but share the same language and reasoning backbone. To the best of our knowledge, this is the first publicly disclosed design of its kind in a unified audio-language model.
37
+ * 📊 **One model, a full audio stack** — FireRedAudio spans ASR, broad and fine-grained audio understanding, zero-shot TTS, Instruct TTS, and free-form speech editing, achieving competitive or leading results across MMAU, MMSU, Seed-TTS-Eval, InstructTTSEval, and Ming-Freeform-Audio-Edit.
38
+ * 🎙️ **Create and edit speech with natural language** — Clone a voice from a reference clip, design a voice from a description, or edit what was said and how it sounds through one continuous-latent generation pathway.
39
+ * ⏱️ **Go from minutes to hour-long recordings** — Understand recordings up to one hour with precise time-to-content alignment. Organize audio into timestamped structures, produce grounded summaries, retrieve content by time (or time by content), and reason over evidence distributed across the recording.
40
+
41
+
42
+ ## News
43
+
44
+ - [2026.08.21] We release the **FireRedAudio** code and model.
45
+
46
+
47
+ ## Contents
48
+
49
+ - [Quick Start](#quick-start-)
50
+ - [Performance](#performance)
51
+ - [Limitations](#limitations)
52
+ - [Usage Disclaimer](#usage-disclaimer)
53
+ - [Citation](#citation)
54
+ - [Acknowledgements](#acknowledgements)
55
+ - [License](#license)
56
+
57
+
58
+ ## Quick Start 🚀
59
+
60
+ ### Installation
61
+
62
+ Requires **Python 3.10** and [uv](https://docs.astral.sh/uv/). System prerequisites: a
63
+ **CUDA toolkit** (GPU inference + compiling the causal-conv1d / flash-attn kernels) and
64
+ **ffmpeg** (audio decoding via torchaudio/torchcodec).
65
+
66
+ Wheels target **CUDA 12.8 (`cu128`)** by default. For a different CUDA version, change the
67
+ `pytorch-cu128` index URL in [`pyproject.toml`](pyproject.toml) (e.g. `cu126` / `cu129`) and
68
+ re-run `uv sync`.
69
+
70
+ ```sh
71
+ uv sync --extra accel --extra accel-build --group tools # + compile causal-conv1d, flash-attn
72
+ ```
73
+
74
+ ### Model Download
75
+
76
+ Download the pretrained model from Hugging Face with the `hf` CLI:
77
+
78
+ ```sh
79
+ hf download FireRedTeam/FireRedAudio --local-dir pretrained_models/
80
+ ```
81
+
82
+ ### Python API
83
+
84
+ ```python
85
+ import torch
86
+ import torchaudio
87
+ from inference import FireRedAudioInference
88
+
89
+ # Init the model. Understanding tasks need only --model; generation tasks
90
+ # additionally need the RedAE decoder weights.
91
+ engine = FireRedAudioInference(
92
+ model_path="pretrained_models/FireRedAudio",
93
+ vae_decoder_path="pretrained_models/RedAE_decoder/model.pt", # only for tts / edit / voice_design
94
+ device="cuda:0",
95
+ )
96
+
97
+ # ---- 1) Speech recognition (ASR) -----------------------------------------
98
+ res = engine.understand("assets/examples/asr_zh_fleurs.wav", "Transcribe speech to text.", task="asr")
99
+ print(res.answer)
100
+
101
+ # ---- 2) Audio understanding (with optional chain-of-thought) --------------
102
+ res = engine.understand(
103
+ "assets/examples/assets_mmau_test.wav",
104
+ "What illness did Second speaker's friend suffer from?\n(A) Progressive arthritis (B) Progressive cancer (C) Acute pneumonia (D) Chronic heart disease",
105
+ task="understand", enable_thinking=True, max_new_tokens=10240,
106
+ )
107
+ print("CoT:")
108
+ print(res.reasoning) # CoT reasoning, or None
109
+ print("Answer")
110
+ print(res.answer)
111
+
112
+ # ---- 3) Zero-shot TTS (ICL voice cloning) ---------------------------------
113
+ res = engine.tts(
114
+ prompt_text="同时,他强调微调要科学有序。",
115
+ prompt_audio="assets/examples/tts_zh_prompt.wav",
116
+ target_text="安徽淮南秦师傅发现,停在小区的爱车右前驾驶窗玻璃被砸。",
117
+ language="zh",
118
+ )
119
+ torchaudio.save("tts.wav", res.audio.cpu(), sample_rate=24000)
120
+
121
+ # ---- 4) Speech editing -----------------------------------------------------
122
+ # semantic: rewrite / substitute / insert / delete content. The model first writes
123
+ # <|sot|>{rewritten text}<|eot|> then renders the audio.
124
+ res = engine.edit("assets/examples/edit_semantic_zh_ref.wav", "delete '比普通的茶叶要'", edit_type="semantic"
125
+ print(res.text)
126
+ torchaudio.save("edit_semantic.wav", res.audio.cpu(), sample_rate=24000)
127
+
128
+ # acoustic: change pitch / speed / volume. The instruction must follow the exact
129
+ # templates below (the model is trained on these, not free-form phrasing):
130
+ # pitch -> "shift the pitch by N step(s)" N in {-6, ..., -1, 1, ..., +6}
131
+ # speed -> "adjust the speed to X" X in [0.5, 2.0], step 0.1
132
+ # volume -> "adjust the volume to X" X in [0.3, 2.0], step 0.1
133
+ res = engine.edit("assets/examples/edit_acoustic_zh_ref.wav", "shift the pitch by 3 steps", edit_type="acoustic")
134
+ torchaudio.save("edit_acoustic.wav", res.audio.cpu(), sample_rate=24000)
135
+
136
+ # ---- 5) Voice design (synthesis from a timbre description) -----------------
137
+ res = engine.voice_design(
138
+ instruction="以女性高音区的清亮音色,表现出青年阶段的特质,音量略强,语速适中稍快,语调带有解释意味和急切的情感流露,确保语音流畅自然。",
139
+ text="是我请他来的,可他什么也不知道,他来只是想打听一下,你们厂是不是有旧锅炉?",
140
+ )
141
+ torchaudio.save("voice_design.wav", res.audio.cpu(), sample_rate=24000)
142
+ ```
143
+
144
+ ### Command Line Inference
145
+
146
+ ```sh
147
+ # speech recognition
148
+ uv run inference.py --task asr --model pretrained_models/FireRedAudio --audio assets/examples/asr_zh_fleurs.wav
149
+
150
+ # audio understanding and QA; several --audio for e.g. speaker verification,
151
+ # --enable-thinking to let the model reason first
152
+ uv run inference.py --task understand --model pretrained_models/FireRedAudio --audio assets/examples/assets_mmau_test.wav \
153
+ --prompt "What illness did Second speaker's friend suffer from?\n(A) Progressive arthritis (B) Progressive cancer (C) Acute pneumonia (D) Chronic heart disease" --enable-thinking --max-new-tokens 4096
154
+
155
+
156
+ # ICL voice cloning from a reference audio and its transcript
157
+ uv run inference.py --task tts --model pretrained_models/FireRedAudio --vae-decoder pretrained_models/RedAE_decoder/model.pt \
158
+ --prompt-audio assets/examples/tts_zh_prompt.wav --prompt-text "同时,他强调微调要科学有序。" \
159
+ --target-text "安徽淮南秦师傅发现,停在小区的爱车右前驾驶窗玻璃被砸。" --language zh --output tts.wav
160
+
161
+ # speech editing; semantic rewrites content, acoustic changes pitch / speed / volume.
162
+ uv run inference.py --task edit --model pretrained_models/FireRedAudio --vae-decoder pretrained_models/RedAE_decoder/model.pt \
163
+ --audio assets/examples/edit_semantic_zh_ref.wav --instruction "delete '比普通的茶叶要'" --edit-type semantic \
164
+ --output edit_semantic.wav
165
+
166
+ # acoustic: instructions must use the exact trained templates, e.g.
167
+ # "shift the pitch by N step(s)" in -6..6 steps (pitch)
168
+ # "adjust the speed to X" in [0.5, 2.0], step .1 (speed)
169
+ # "adjust the volume to X" in [0.3, 2.0], step .1 (volume)
170
+ uv run inference.py --task edit --model pretrained_models/FireRedAudio --vae-decoder pretrained_models/RedAE_decoder/model.pt \
171
+ --audio assets/examples/edit_acoustic_zh_ref.wav --instruction "shift the pitch by 3 steps" --edit-type acoustic \
172
+ --output edit_acoustic.wav
173
+
174
+ # voice design
175
+ uv run inference.py --task voice_design --model pretrained_models/FireRedAudio --vae-decoder pretrained_models/RedAE_decoder/model.pt \
176
+ --instruction "以女性高音区的清亮音色,表现出青年阶段的特质,音量略强,语速适中稍快,语调带有解释意味和急切的情感流露,确保语音流畅自然。" \
177
+ --text "是我请他来的,可他什么也不知道,他来只是想打听一下,你们厂是不是有旧锅炉?" --output voice_design.wav
178
+ ```
179
+
180
+
181
+ ## Performance
182
+
183
+ ### Audio Understanding
184
+
185
+ <div align="center">
186
+
187
+ <table style="border-collapse:collapse; margin:0 auto; text-align:center; white-space:nowrap;">
188
+ <thead>
189
+ <tr>
190
+ <th align="center" style="padding:6px 14px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">Model</div></th>
191
+ <th align="center" style="padding:6px 14px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">MMAU test-mini</div></th>
192
+ <th align="center" style="padding:6px 14px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">MMAU test</div></th>
193
+ <th align="center" style="padding:6px 14px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">MMSU</div></th>
194
+ </tr>
195
+ </thead>
196
+ <tbody>
197
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">Step-Audio-R1.1</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">77.7</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">75.9</td></tr>
198
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">Step-Audio 2</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">78.0</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">–</td></tr>
199
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">MiMo-Audio-7B-Instruct</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">74.9</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">61.7</td></tr>
200
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">Kimi-Audio</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">65.2</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">–</td></tr>
201
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">LongCat-Next</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">76.4</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">–</td></tr>
202
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">Qwen3-Omni-30B-A3B-Instruct</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">77.5</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">69.0</td></tr>
203
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">Gemini 3.1 Pro</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">80.7*</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">78.8*</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">82.7*</td></tr>
204
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">Qwen3.5-Omni-Plus</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">81.4*</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">79.9*</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">80.7*</td></tr>
205
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;"><b>FireRedAudio</b></td><td align="center" style="padding:6px 14px; border:1px solid #ddd;"><b>82.0</b></td><td align="center" style="padding:6px 14px; border:1px solid #ddd;"><b>80.9</b></td><td align="center" style="padding:6px 14px; border:1px solid #ddd;"><b>83.3</b></td></tr>
206
+ </tbody>
207
+ </table>
208
+
209
+ </div>
210
+
211
+ * *Results marked with * are obtained from our own evaluation.*
212
+
213
+ ### ASR
214
+
215
+ <div align="center">
216
+
217
+ <table style="border-collapse:collapse; margin:0 auto; text-align:center; white-space:nowrap;">
218
+ <thead>
219
+ <tr>
220
+ <th align="center" style="padding:6px 12px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">Model</div></th>
221
+ <th align="center" style="padding:6px 12px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">AISHELL&#8209;1</div></th>
222
+ <th align="center" style="padding:6px 12px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">AISHELL&#8209;2<br>test&#8209;ios</div></th>
223
+ <th align="center" style="padding:6px 12px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">WenetSpeech<br>Net&nbsp;&#8288;|&#8288;&nbsp;Meeting</div></th>
224
+ <th align="center" style="padding:6px 12px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">LibriSpeech<br>clean&nbsp;&#8288;|&#8288;&nbsp;other</div></th>
225
+ <th align="center" style="padding:6px 12px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">FLEURS<br>en&nbsp;&#8288;|&#8288;&nbsp;zh</div></th>
226
+ <th align="center" style="padding:6px 12px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">FLEURS&#8209;102<br>avg</div></th>
227
+ <th align="center" style="padding:6px 12px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">KeSpeech</div></th>
228
+ <th align="center" style="padding:6px 12px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">Opencpop</div></th>
229
+ </tr>
230
+ </thead>
231
+ <tbody>
232
+ <tr><td align="center" style="padding:6px 12px; border:1px solid #ddd;">Step&#8209;Audio&nbsp;2</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">0.63</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">2.10</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">4.67&nbsp;&#8288;|&#8288;&nbsp;4.75</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">1.17&nbsp;&#8288;|&#8288;&nbsp;2.42</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">3.03&nbsp;&#8288;|&#8288;&nbsp;2.68</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">3.63</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td></tr>
233
+ <tr><td align="center" style="padding:6px 12px; border:1px solid #ddd;">MiMo&#8209;Audio&#8209;7B&#8209;Instruct</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">1.65</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">3.50&nbsp;&#8288;|&#8288;&nbsp;–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td></tr>
234
+ <tr><td align="center" style="padding:6px 12px; border:1px solid #ddd;">Ming&#8209;UniAudio&#8209;16B&#8209;A3B</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">2.84</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">1.62&nbsp;&#8288;|&#8288;&nbsp;–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td></tr>
235
+ <tr><td align="center" style="padding:6px 12px; border:1px solid #ddd;">Kimi&#8209;Audio</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">0.60</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">2.56</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">5.37&nbsp;&#8288;|&#8288;&nbsp;6.28</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">1.28&nbsp;&#8288;|&#8288;&nbsp;2.42</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">4.44&nbsp;&#8288;|&#8288;&nbsp;2.69</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td></tr>
236
+ <tr><td align="center" style="padding:6px 12px; border:1px solid #ddd;">LongCat&#8209;Next</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">1.47</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">2.82</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">5.98&nbsp;&#8288;|&#8288;&nbsp;8.19</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">1.63&nbsp;&#8288;|&#8288;&nbsp;3.42</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">5.24&nbsp;&#8288;|&#8288;&nbsp;3.24</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td></tr>
237
+ <tr><td align="center" style="padding:6px 12px; border:1px solid #ddd;">Qwen3&#8209;Omni&#8209;30B&#8209;A3B&#8209;Instruct</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">4.69&nbsp;&#8288;|&#8288;&nbsp;5.89</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">1.22&nbsp;&#8288;|&#8288;&nbsp;2.48</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">2.72&nbsp;&#8288;|&#8288;&nbsp;2.20</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">–</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">1.54</td></tr>
238
+ <tr><td align="center" style="padding:6px 12px; border:1px solid #ddd;">Gemini&nbsp;3.1&nbsp;Pro</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">3.66*</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">7.10*</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">11.53&nbsp;&#8288;|&#8288;&nbsp;14.21</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">3.36&nbsp;&#8288;|&#8288;&nbsp;4.41</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">2.97*&nbsp;&#8288;|&#8288;&nbsp;4.28*</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">18.23*</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">23.67</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">6.83</td></tr>
239
+ <tr><td align="center" style="padding:6px 12px; border:1px solid #ddd;">Qwen3.5&#8209;Omni&#8209;Plus</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">0.82*</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">2.26*</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">4.30&nbsp;&#8288;|&#8288;&nbsp;5.84</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">1.11&nbsp;&#8288;|&#8288;&nbsp;2.23</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">3.33*&nbsp;&#8288;|&#8288;&nbsp;2.46*</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">23.66*</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">3.46</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">1.49</td></tr>
240
+ <tr><td align="center" style="padding:6px 12px; border:1px solid #ddd;"><b>FireRedAudio</b></td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">0.71</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">2.63</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">5.18&nbsp;&#8288;|&#8288;&nbsp;5.33</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">0.67&nbsp;&#8288;|&#8288;&nbsp;2.91</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">2.53&nbsp;&#8288;|&#8288;&nbsp;3.14</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">14.94</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">4.82</td><td align="center" style="padding:6px 12px; border:1px solid #ddd;">1.63</td></tr>
241
+ </tbody>
242
+ </table>
243
+
244
+ </div>
245
+
246
+ * *Results marked with * are obtained from our own evaluation.*
247
+
248
+ ### Zero-Shot TTS
249
+
250
+ <div align="center">
251
+
252
+ <table align="center" style="border-collapse:collapse; margin:0 auto; text-align:center; white-space:nowrap;">
253
+ <thead>
254
+ <tr>
255
+ <th align="center" style="padding:6px 14px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">Model</div></th>
256
+ <th align="center" style="padding:6px 14px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">Seed-ZH<br>CER↓&nbsp;&#8288;|&#8288;&nbsp;SIM↑</div></th>
257
+ <th align="center" style="padding:6px 14px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">Seed-EN<br>WER↓&nbsp;&#8288;|&#8288;&nbsp;SIM↑</div></th>
258
+ <th align="center" style="padding:6px 14px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">Avg.<br>CER/WER↓&nbsp;&#8288;|&#8288;&nbsp;SIM↑</div></th>
259
+ </tr>
260
+ </thead>
261
+ <tbody>
262
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">Seed-TTS</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.12&nbsp;&#8288;|&#8288;&nbsp;<b>0.80</b></td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">2.25&nbsp;&#8288;|&#8288;&nbsp;<b>0.76</b></td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.69&nbsp;&#8288;|&#8288;&nbsp;<b>0.78</b></td></tr>
263
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">FireRedTTS</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.51&nbsp;&#8288;|&#8288;&nbsp;0.65</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">3.82&nbsp;&#8288;|&#8288;&nbsp;0.53</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">2.67&nbsp;&#8288;|&#8288;&nbsp;0.59</td></tr>
264
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">FireRedTTS-2</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.14&nbsp;&#8288;|&#8288;&nbsp;0.74</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.95&nbsp;&#8288;|&#8288;&nbsp;0.65</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.55&nbsp;&#8288;|&#8288;&nbsp;0.69</td></tr>
265
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">DiTAR (1B)</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.02&nbsp;&#8288;|&#8288;&nbsp;0.75</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.69&nbsp;&#8288;|&#8288;&nbsp;<ins>0.74</ins></td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.36&nbsp;&#8288;|&#8288;&nbsp;<ins>0.75</ins></td></tr>
266
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">F5-TTS</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.56&nbsp;&#8288;|&#8288;&nbsp;0.74</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.83&nbsp;&#8288;|&#8288;&nbsp;0.65</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.70&nbsp;&#8288;|&#8288;&nbsp;0.70</td></tr>
267
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">CosyVoice 2</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.45&nbsp;&#8288;|&#8288;&nbsp;0.75</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">2.57&nbsp;&#8288;|&#8288;&nbsp;0.65</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">2.01&nbsp;&#8288;|&#8288;&nbsp;0.70</td></tr>
268
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">CosyVoice 3-1.5B</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.12&nbsp;&#8288;|&#8288;&nbsp;<ins>0.78</ins></td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">2.21&nbsp;&#8288;|&#8288;&nbsp;0.72</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.67&nbsp;&#8288;|&#8288;&nbsp;<ins>0.75</ins></td></tr>
269
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">MiMo-Audio-7B-Instruct</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.96&nbsp;&#8288;|&#8288;&nbsp;–</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">5.37&nbsp;&#8288;|&#8288;&nbsp;–</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">3.67&nbsp;&#8288;|&#8288;&nbsp;–</td></tr>
270
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">Qwen2.5-Omni-7B (RL)</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.42&nbsp;&#8288;|&#8288;&nbsp;0.75</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">2.33&nbsp;&#8288;|&#8288;&nbsp;0.64</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.88&nbsp;&#8288;|&#8288;&nbsp;0.70</td></tr>
271
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">Qwen3-Omni-30B-A3B-Instruct</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.07&nbsp;&#8288;|&#8288;&nbsp;–</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;"><b>1.39</b>&nbsp;&#8288;|&#8288;&nbsp;–</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;"><ins>1.23</ins>&nbsp;&#8288;|&#8288;&nbsp;–</td></tr>
272
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">Ming-UniAudio-16B-A3B</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;"><ins>0.95</ins>&nbsp;&#8288;|&#8288;&nbsp;0.70</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.85&nbsp;&#8288;|&#8288;&nbsp;0.58</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">1.40&nbsp;&#8288;|&#8288;&nbsp;0.64</td></tr>
273
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;"><b>FireRedAudio</b></td><td align="center" style="padding:6px 14px; border:1px solid #ddd;"><b>0.83</b>&nbsp;&#8288;|&#8288;&nbsp;0.74</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;"><ins>1.56</ins>&nbsp;&#8288;|&#8288;&nbsp;0.68</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;"><b>1.20</b>&nbsp;&#8288;|&#8288;&nbsp;0.71</td></tr>
274
+ </tbody>
275
+ </table>
276
+
277
+ </div>
278
+
279
+ ### Instruct TTS
280
+
281
+ <div align="center">
282
+
283
+ <table style="border-collapse:collapse; margin:0 auto; text-align:center; white-space:nowrap;">
284
+ <thead>
285
+ <tr>
286
+ <th align="center" style="padding:6px 14px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">Model</div></th>
287
+ <th align="center" style="padding:6px 14px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">ZH<br>APS↑&nbsp;&#8288;|&#8288;&nbsp;DSD↑&nbsp;&#8288;|&#8288;&nbsp;RP↑</div></th>
288
+ <th align="center" style="padding:6px 14px; border:1px solid #ddd; background-color:#f5f5f5;"><div align="center">EN<br>APS↑&nbsp;&#8288;|&#8288;&nbsp;DSD↑&nbsp;&#8288;|&#8288;&nbsp;RP↑</div></th>
289
+ </tr>
290
+ </thead>
291
+ <tbody>
292
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">VoiceSculptor-VD</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">74.6&nbsp;&#8288;|&#8288;&nbsp;63.5&nbsp;&#8288;|&#8288;&nbsp;62.0</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">–&nbsp;&#8288;|&#8288;&nbsp;–&nbsp;&#8288;|&#8288;&nbsp;–</td></tr>
293
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">MOSS-VoiceGenerator</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">71.6&nbsp;&#8288;|&#8288;&nbsp;72.5&nbsp;&#8288;|&#8288;&nbsp;61.3</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">58.8&nbsp;&#8288;|&#8288;&nbsp;71.8&nbsp;&#8288;|&#8288;&nbsp;61.6</td></tr>
294
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">Ming-Omni-TTS-16B</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">84.6&nbsp;&#8288;|&#8288;&nbsp;70.7&nbsp;&#8288;|&#8288;&nbsp;56.0</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">–&nbsp;&#8288;|&#8288;&nbsp;–&nbsp;&#8288;|&#8288;&nbsp;–</td></tr>
295
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;">Qwen3-TTS-VD</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">83.7&nbsp;&#8288;|&#8288;&nbsp;81.7&nbsp;&#8288;|&#8288;&nbsp;65.8</td><td align="center" style="padding:6px 14px; border:1px solid #ddd;">76.4&nbsp;&#8288;|&#8288;&nbsp;81.4&nbsp;&#8288;|&#8288;&nbsp;64.2</td></tr>
296
+ <tr><td align="center" style="padding:6px 14px; border:1px solid #ddd;"><b>FireRedAudio</b></td><td align="center" style="padding:6px 14px; border:1px solid #ddd;"><b>86.0</b>&nbsp;&#8288;|&#8288;&nbsp;<b>84.1</b>&nbsp;&#8288;|&#8288;&nbsp;<b>70.1</b></td><td align="center" style="padding:6px 14px; border:1px solid #ddd;"><b>81.1</b>&nbsp;&#8288;|&#8288;&nbsp;<b>83.6</b>&nbsp;&#8288;|&#8288;&nbsp;<b>70.3</b></td></tr>
297
+ </tbody>
298
+ </table>
299
+
300
+ </div>
301
+
302
+ * *These results are obtained from our own evaluation.*
303
+
304
+ ### Semantic Editing
305
+
306
+ <div align="center">
307
+
308
+ <table align="center" style="border-collapse:collapse; margin:0 auto; text-align:center; white-space:nowrap;">
309
+ <thead>
310
+ <tr>
311
+ <th align="center" style="padding:4px 12px; border:1px solid #ddd;"><div align="center">Task</div></th>
312
+ <th align="center" style="padding:4px 12px; border:1px solid #ddd;"><div align="center">Setting</div></th>
313
+ <th align="center" style="padding:4px 12px; border:1px solid #ddd;"><div align="center">Metric</div></th>
314
+ <th align="center" style="padding:4px 12px; border:1px solid #ddd;"><div align="center">Ming-UniAudio-Edit<br>zh | en</div></th>
315
+ <th align="center" style="padding:4px 12px; border:1px solid #ddd;"><div align="center">FireRedAudio<br>zh | en</div></th>
316
+ </tr>
317
+ </thead>
318
+ <tbody>
319
+ <tr>
320
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;" rowspan="8"><b>Deletion</b></td>
321
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;" rowspan="4"><b>basic</b></td>
322
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">WER (%)↓</td>
323
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">11.89 | 14.85</td>
324
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>10.82</b> | <b>12.78</b></td>
325
+ </tr>
326
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">SIM↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>0.78</b> | 0.76</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>0.78</b> | <b>0.79</b></td></tr>
327
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">ACC (%)↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>100.00</b> | 82.22</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>100.00</b> | <b>97.78</b></td></tr>
328
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">no-edit WER (%)↓</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">11.49 | 24.26</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>10.70</b> | <b>23.16</b></td></tr>
329
+ <tr>
330
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;" rowspan="4"><b>open</b></td>
331
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">WER (%)↓</td>
332
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">22.92 | 27.60</td>
333
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>10.49</b> | <b>16.65</b></td>
334
+ </tr>
335
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">SIM↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>0.81</b> | 0.74</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">0.80 | <b>0.80</b></td></tr>
336
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">ACC (%)↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">82.92 | 85.00</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>89.32</b> | <b>86.50</b></td></tr>
337
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">no-edit WER (%)↓</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">17.50 | 35.21</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>7.84</b> | <b>25.43</b></td></tr>
338
+ <tr>
339
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;" rowspan="8"><b>Insertion</b></td>
340
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;" rowspan="4"><b>basic</b></td>
341
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">WER (%)↓</td>
342
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">3.42 | 6.63</td>
343
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>3.28</b> | <b>4.98</b></td>
344
+ </tr>
345
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">SIM↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>0.83</b> | 0.79</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>0.83</b> | <b>0.84</b></td></tr>
346
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">ACC (%)↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">80.00 | 71.43</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>83.53</b> | <b>87.58</b></td></tr>
347
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">no-edit WER (%)↓</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">3.52 | 17.70</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>3.51</b> | <b>16.56</b></td></tr>
348
+ <tr>
349
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;" rowspan="4"><b>open</b></td>
350
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">WER (%)↓</td>
351
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">3.89 | 7.59</td>
352
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>2.57</b> | <b>6.98</b></td>
353
+ </tr>
354
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">SIM↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>0.83</b> | 0.79</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>0.83</b> | <b>0.84</b></td></tr>
355
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">ACC (%)↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">79.31 | 62.31</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>86.90</b> | <b>69.85</b></td></tr>
356
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">no-edit WER (%)↓</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">4.10 | 18.84</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>2.77</b> | <b>17.83</b></td></tr>
357
+ <tr>
358
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;" rowspan="8"><b>Substitution</b></td>
359
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;" rowspan="4"><b>basic</b></td>
360
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">WER (%)↓</td>
361
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">4.52 | 8.99</td>
362
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>2.66</b> | <b>4.46</b></td>
363
+ </tr>
364
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">SIM↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">0.82 | 0.78</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>0.84</b> | <b>0.81</b></td></tr>
365
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">ACC (%)↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">78.62 | 59.78</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>87.42</b> | <b>75.98</b></td></tr>
366
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">no-edit WER (%)↓</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">4.63 | 19.28</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>2.91</b> | <b>16.34</b></td></tr>
367
+ <tr>
368
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;" rowspan="4"><b>open</b></td>
369
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">WER (%)↓</td>
370
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">4.56 | 7.64</td>
371
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>2.45</b> | <b>4.41</b></td>
372
+ </tr>
373
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">SIM↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">0.83 | 0.77</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>0.84</b> | <b>0.81</b></td></tr>
374
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">ACC (%)↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">76.62 | 65.62</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>90.15</b> | <b>76.95</b></td></tr>
375
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">no-edit WER (%)↓</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">4.75 | 18.39</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>2.71</b> | <b>16.16</b></td></tr>
376
+ </tbody>
377
+ </table>
378
+
379
+ </div>
380
+
381
+ ### Acoustic Editing
382
+
383
+
384
+ <div align="center">
385
+
386
+ <table style="border-collapse:collapse; margin:0 auto; text-align:center; white-space:nowrap;">
387
+ <thead>
388
+ <tr>
389
+ <th align="center" style="padding:4px 12px; border:1px solid #ddd;"><div align="center">Task</div></th>
390
+ <th align="center" style="padding:4px 12px; border:1px solid #ddd;"><div align="center">Metric</div></th>
391
+ <th align="center" style="padding:4px 12px; border:1px solid #ddd;"><div align="center">Ming-UniAudio-Edit<br>zh | en</div></th>
392
+ <th align="center" style="padding:4px 12px; border:1px solid #ddd;"><div align="center">FireRedAudio<br>zh | en</div></th>
393
+ </tr>
394
+ </thead>
395
+ <tbody>
396
+ <tr>
397
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;" rowspan="3"><b>Speed Alteration</b></td>
398
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">WER (%)↓</td>
399
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">5.88 | 17.53</td>
400
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>2.00</b> | <b>4.43</b></td>
401
+ </tr>
402
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">SIM↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">0.66 | 0.57</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>0.79</b> | <b>0.71</b></td></tr>
403
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">RDE (%)↓</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">6.36 | 5.92</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>2.60</b> | <b>4.02</b></td></tr>
404
+ <tr>
405
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;" rowspan="2"><b>Pitch Alteration</b></td>
406
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">WER (%)↓</td>
407
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">7.45 | 13.37</td>
408
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>2.00</b> | <b>3.04</b></td>
409
+ </tr>
410
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">SIM↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">0.36 | 0.24</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>0.52</b> | <b>0.44</b></td></tr>
411
+ <tr>
412
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;" rowspan="3"><b>Volume Alteration</b></td>
413
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">WER (%)↓</td>
414
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;">1.71 | 1.35</td>
415
+ <td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>1.60</b> | <b>1.30</b></td>
416
+ </tr>
417
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">SIM↑</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">0.86 | 0.80</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>0.94</b> | <b>0.93</b></td></tr>
418
+ <tr><td align="center" style="padding:4px 12px; border:1px solid #ddd;">RAE (%)↓</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;">14.90 | 11.70</td><td align="center" style="padding:4px 12px; border:1px solid #ddd;"><b>2.39</b> | <b>3.74</b></td></tr>
419
+ </tbody>
420
+ </table>
421
+
422
+ </div>
423
+
424
+
425
+
426
+ ## Limitations
427
+
428
+ - **Everything but ASR is Chinese/English only.** Speech generation (`tts` / `edit` / `voice_design`) and audio understanding are limited to Chinese and English — `tts` selects the language via `--language zh` / `en`. ASR is the only task that supports more languages.
429
+ - **Zero-shot TTS is not deterministic by default.** The flow-matching decoder samples random noise, so output varies run to run; pass a fixed seed (`set_seed(...)` in the API, `--seed` on the CLI) for reproducibility, and note that quality can differ across seeds.
430
+ - **Long-form input is supported up to about one hour.** Beyond that the model is untested and time-to-content alignment may degrade.
431
+
432
+
433
+ ## Usage Disclaimer
434
+
435
+ - The project incorporates zero-shot voice cloning functionality; Please note that this capability is intended **solely for academic research purposes**.
436
+ - **DO NOT** use this model for **ANY illegal activities**❗️❗️
437
+ - The developers assume no liability for any misuse of this model.
438
+ - If you identify any instances of **abuse**, **misuse**, or **fraudulent** activities related to this project, **please report them to our team immediately.**
439
+
440
+
441
+ ## Citation
442
+
443
+ ```bib
444
+ @article{fireredaudio,
445
+ title = {FireRedAudio: A General-Purpose Audio Language Model with Decoupled Continuous Representations for Understanding and Generation},
446
+ author = {FireRed Team},
447
+ journal = {arXiv preprint},
448
+ year = {2026},
449
+ }
450
+ ```
451
+
452
+
453
+ ## Acknowledgements
454
+
455
+ - [Qwen3.5](https://github.com/QwenLM/Qwen3.8) for the language model foundation
456
+ - [Whisper-large-v3](https://huggingface.co/openai/whisper-large-v3) for the Audio Encoder initialization
457
+ - [x-transformers](https://github.com/lucidrains/x-transformers) for RotaryEmbedding
458
+ - [vocos](https://github.com/gemelo-ai/vocos/tree/main) for ISTFT implementation
459
+
460
+
461
+ ## License
462
+
463
+ Released under the [Apache-2.0](LICENSE) license.