SOTA-Class TTS at Compact Scale
Python
Expand evaluation parameter comparisons
Your Nameeb14a3a4cbb5
diff --git a/README.md b/README.mdindex 4dec549..2fa244c 100644--- a/README.md+++ b/README.md@@ -183,31 +183,46 @@ standard `AutoModel` and `AutoProcessor` APIs using `trust_remote_code=True`.## Evaluation+Audio8 TTS Preview is the smallest model in this comparison at just **0.6B+parameters**. Despite using only a fraction of the parameters of the other+systems, it delivers results in the first tier of industry-leading SOTA TTS+models on the benchmarks below. In particular, it achieves the best English+WER and competitive Chinese CER on Seed-TTS, while remaining competitive+across the CV3 multilingual evaluation.+Lower WER/CER is better; higher SIM is better. Seed-TTS similarity values areshown as percentages.### Seed-TTS-| Model | EN WER / SIM | ZH CER / SIM | Hard ZH CER / SIM |-|---|---:|---:|---:|-| audio8_tts Preview | 1.506 / 63.2 | 0.950 / 73.1 | 11.510 / 68.7 |-| Fish S2 Pro | 1.607 / 64.6 | 1.038 / 73.8 | 10.149 / 70.1 |-| Higgs Audio v2 | 1.524 / 66.4 | 0.806 / 72.1 | 10.622 / 69.3 |-| CosyVoice3-1.5B | 2.22 / 72.0 | 1.12 / 78.1 | 5.83 / 75.8 |-| MOSS-TTS | 1.85 / 73.4 | 1.20 / 78.8 | - |-| VoxCPM2 | 1.84 / 75.3 | 0.97 / 79.5 | 8.13 / 75.3 |+| Model | Parameters | EN WER / SIM | ZH CER / SIM | Hard ZH CER / SIM |+|---|---:|---:|---:|---:|+| **Audio8 TTS Preview** | **0.6B** | **1.506** / 63.2 | 0.950 / 73.1 | 11.510 / 68.7 |+| Fish S2 Pro | 4.6B | 1.607 / 64.6 | 1.038 / 73.8 | 10.149 / 70.1 |+| Higgs Audio v2 | 4.7B | 1.524 / 66.4 | **0.806** / 72.1 | 10.622 / 69.3 |+| CosyVoice3-1.5B | 1.5B | 2.22 / 72.0 | 1.12 / 78.1 | **5.83** / **75.8** |+| MOSS-TTS | 8.5B | 1.85 / 73.4 | 1.20 / 78.8 | - |+| VoxCPM2 | 2.3B | 1.84 / **75.3** | 0.97 / **79.5** | 8.13 / 75.3 |### CV3 multilingual error rate-| Model | zh | en | hard-zh | hard-en | ja | ko | de | es | fr | it | ru |-|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|-| **audio8_tts Preview** | 3.205 | 3.128 | 10.535 | 5.997 | 7.205 | 4.223 | 3.447 | 3.641 | 8.790 | 4.790 | - |-| Fish S2 Pro | 3.600 | 3.493 | 10.588 | 7.349 | 5.139 | 4.111 | 3.605 | 2.972 | 8.600 | 4.229 | 4.702 |-| Higgs Audio v2 | 3.378 | 3.404 | 10.424 | 5.754 | 4.742 | 4.260 | 3.300 | 2.929 | 9.425 | 3.555 | 5.423 |-| CosyVoice3-1.5B | 3.91 | 4.99 | 9.77 | 10.55 | 7.57 | 5.69 | 6.43 | 4.47 | 11.8 | 10.5 | 6.64 |-| VoxCPM2 | 3.65 | 5.00 | 8.55 | 8.48 | 5.96 | 5.69 | 4.77 | 3.80 | 9.85 | 4.25 | 5.21 |--Fish S2 Pro was reevaluated because its official evaluation uses its own normalizer, Higgs Audio v2 was evaluated locally because concrete values were unavailable, and all other baselines use their official values as collected by the [VoxCPM repository](https://github.com/OpenBMB/VoxCPM).+| Model | Parameters | zh | en | hard-zh | hard-en | ja | ko | de | es | fr | it | ru |+|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|+| **Audio8 TTS Preview** | **0.6B** | **3.205** | **3.128** | 10.535 | 5.997 | 7.205 | 4.223 | 3.447 | 3.641 | 8.790 | 4.790 | - |+| Fish S2 Pro | 4.6B | 3.600 | 3.493 | 10.588 | 7.349 | 5.139 | **4.111** | 3.605 | 2.972 | **8.600** | 4.229 | **4.702** |+| Higgs Audio v2 | 4.7B | 3.378 | 3.404 | 10.424 | **5.754** | **4.742** | 4.260 | **3.300** | **2.929** | 9.425 | **3.555** | 5.423 |+| CosyVoice3-1.5B | 1.5B | 3.91 | 4.99 | 9.77 | 10.55 | 7.57 | 5.69 | 6.43 | 4.47 | 11.8 | 10.5 | 6.64 |+| VoxCPM2 | 2.3B | 3.65 | 5.00 | **8.55** | 8.48 | 5.96 | 5.69 | 4.77 | 3.80 | 9.85 | 4.25 | 5.21 |++Parameter counts are calculated directly from the released weight tensors.+MOSS-TTS contains 8,489,841,664 parameters. VoxCPM2's main model contains+2,290,004,544 parameters; the separate AudioVAE is not included in the+parameter comparison.++Fish S2 Pro was reevaluated because its official evaluation uses its own+normalizer. Higgs Audio v2 was evaluated locally because concrete values were+unavailable. All other baseline values were collected from their official+reports through the [VoxCPM repository](https://github.com/OpenBMB/VoxCPM).Different normalizers and evaluators make cross-project values referencecomparisons rather than a strictly matched ranking. Evaluation coverage doesdiff --git a/README_zh.md b/README_zh.mdindex 7b2eda7..2a4cb2e 100644--- a/README_zh.md+++ b/README_zh.md@@ -169,30 +169,39 @@ SFT 同时优化 slow semantic/EOS loss 和 fast codebook teacher-forcing loss## 评测结果+Audio8 TTS Preview 仅有 **0.6B 参数**,是本次对比中规模最小的模型。尽管参数量仅为其他模型的+一小部分,它在以下基准测试中依然达到业界领先 TTS 模型的第一梯队水平。其中,Audio8 TTS+Preview 在 Seed-TTS 上取得了最低的英文 WER 和有竞争力的中文 CER,并在 CV3 多语言评测中+保持了具有竞争力的整体表现。+WER/CER 越低越好,SIM 越高越好。Seed-TTS 的相似度统一显示为百分数。### Seed-TTS-| 模型 | EN WER / SIM | ZH CER / SIM | Hard ZH CER / SIM |-|---|---:|---:|---:|-| audio8_tts Preview | 1.506 / 63.2 | 0.950 / 73.1 | 11.510 / 68.7 |-| Fish S2 Pro | 1.607 / 64.6 | 1.038 / 73.8 | 10.149 / 70.1 |-| Higgs Audio v2 | 1.524 / 66.4 | 0.806 / 72.1 | 10.622 / 69.3 |-| CosyVoice3-1.5B | 2.22 / 72.0 | 1.12 / 78.1 | 5.83 / 75.8 |-| MOSS-TTS | 1.85 / 73.4 | 1.20 / 78.8 | - |-| VoxCPM2 | 1.84 / 75.3 | 0.97 / 79.5 | 8.13 / 75.3 |+| 模型 | 参数量 | EN WER / SIM | ZH CER / SIM | Hard ZH CER / SIM |+|---|---:|---:|---:|---:|+| **Audio8 TTS Preview** | **0.6B** | **1.506** / 63.2 | 0.950 / 73.1 | 11.510 / 68.7 |+| Fish S2 Pro | 4.6B | 1.607 / 64.6 | 1.038 / 73.8 | 10.149 / 70.1 |+| Higgs Audio v2 | 4.7B | 1.524 / 66.4 | **0.806** / 72.1 | 10.622 / 69.3 |+| CosyVoice3-1.5B | 1.5B | 2.22 / 72.0 | 1.12 / 78.1 | **5.83** / **75.8** |+| MOSS-TTS | 8.5B | 1.85 / 73.4 | 1.20 / 78.8 | - |+| VoxCPM2 | 2.3B | 1.84 / **75.3** | 0.97 / **79.5** | 8.13 / 75.3 |### CV3 多语言错误率-| 模型 | zh | en | hard-zh | hard-en | ja | ko | de | es | fr | it | ru |-|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|-| **audio8_tts Preview** | 3.205 | 3.128 | 10.535 | 5.997 | 7.205 | 4.223 | 3.447 | 3.641 | 8.790 | 4.790 | - |-| Fish S2 Pro | 3.600 | 3.493 | 10.588 | 7.349 | 5.139 | 4.111 | 3.605 | 2.972 | 8.600 | 4.229 | 4.702 |-| Higgs Audio v2 | 3.378 | 3.404 | 10.424 | 5.754 | 4.742 | 4.260 | 3.300 | 2.929 | 9.425 | 3.555 | 5.423 |-| CosyVoice3-1.5B | 3.91 | 4.99 | 9.77 | 10.55 | 7.57 | 5.69 | 6.43 | 4.47 | 11.8 | 10.5 | 6.64 |-| VoxCPM2 | 3.65 | 5.00 | 8.55 | 8.48 | 5.96 | 5.69 | 4.77 | 3.80 | 9.85 | 4.25 | 5.21 |+| 模型 | 参数量 | zh | en | hard-zh | hard-en | ja | ko | de | es | fr | it | ru |+|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|+| **Audio8 TTS Preview** | **0.6B** | **3.205** | **3.128** | 10.535 | 5.997 | 7.205 | 4.223 | 3.447 | 3.641 | 8.790 | 4.790 | - |+| Fish S2 Pro | 4.6B | 3.600 | 3.493 | 10.588 | 7.349 | 5.139 | **4.111** | 3.605 | 2.972 | **8.600** | 4.229 | **4.702** |+| Higgs Audio v2 | 4.7B | 3.378 | 3.404 | 10.424 | **5.754** | **4.742** | 4.260 | **3.300** | **2.929** | 9.425 | **3.555** | 5.423 |+| CosyVoice3-1.5B | 1.5B | 3.91 | 4.99 | 9.77 | 10.55 | 7.57 | 5.69 | 6.43 | 4.47 | 11.8 | 10.5 | 6.64 |+| VoxCPM2 | 2.3B | 3.65 | 5.00 | **8.55** | 8.48 | 5.96 | 5.69 | 4.77 | 3.80 | 9.85 | 4.25 | 5.21 |++参数量直接根据已发布的权重张量统计。MOSS-TTS 包含 8,489,841,664 个参数;VoxCPM2 主模型+包含 2,290,004,544 个参数,单独的 AudioVAE 不计入本次参数量对比。-Fish S2 Pro 因官方评测使用自身 normalizer 而重新评测,Higgs Audio v2 因未公布具体值而自行评测,其他模型均采用 [VoxCPM 官方仓库](https://github.com/OpenBMB/VoxCPM)汇总的各项目官方值。+Fish S2 Pro 因官方评测使用自身 normalizer 而重新评测。Higgs Audio v2 因未公布具体数值而自行+评测。其他基线均采用 [VoxCPM 官方仓库](https://github.com/OpenBMB/VoxCPM)汇总的各项目官方值。不同 normalizer 和评测器下的跨项目数值只能作为参考,不能视为严格同口径排名;评测覆盖也不会将 Preview 的正式支持范围扩展到前述 11 种语言之外。