ZONOS2
Official repository and model card - 2026
Official ZONOS2 resources document the Apache-2.0 release, multilingual TTS API, more than 6M hours of training speech, high-fidelity voice cloning, and 44.1 kHz PCM output.
Loading...
Input
Reference voice audio plus multilingual text for ZONOS2 voice cloning TTS
Audio formats
Output
Generate 44.1 kHz WAV speech
Use case
Short-reference narration, voice tests, and multilingual speech samples
Include the tool name, steps and the problem in your feedback. Screenshots can help us locate it.
ZONOS2 Voice Cloning TTS generates multilingual speech from a short reference voice clip. TelkNet presents it as a server-side official ZONOS2 deployment with highest-quality defaults and a simple upload-text-download workflow.
ZONOS2 is Zyphra's Apache-2.0 8B total / 900M active MoE TTS model for multilingual zero-shot voice cloning and 44.1 kHz PCM output.
ZONOS2 returns 44.1 kHz PCM audio through its DAC codec path.
This checkpoint has no reproducible public score table. See the linked release material for details.
Official repository and model card - 2026
Official ZONOS2 resources document the Apache-2.0 release, multilingual TTS API, more than 6M hours of training speech, high-fidelity voice cloning, and 44.1 kHz PCM output.
No paper links are provided.
Includes public test results for models used here and elsewhere. The complete source table is shown by default; conditions and metrics can also be selected for comparison. Scores from different test conditions are not directly comparable.
Only records with verified public inference implementations and matching weights are shown; algorithms requiring no weights are also eligible. Closed services, unreleased weights and unmatched experiment versions are excluded. This is not an all-model leaderboard.
Public weights do not establish a fully open-source license. Expand the data for declared weight licenses and access conditions. Undeclared licenses are marked separately.
The selected model has no verified matching public result and has not been ranked.
Choose a dataset to view all metrics for each model. Rows follow the official rank or selected metric. Bold marks each column’s best comparable value, including ties. Different evaluation protocols are ranked separately; missing scores are shown as -.
Deployable entries 6 / source rows 17
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
CosyVoice 3 · DiTAR · F5-TTS · FireRedTTS-2 · IndexTTS 2 · MegaTTS 3 · MiniMax-Speech · Qwen3-TTS · Seed-TTS · VibeVoice · dots.tts (MF-1step)
Values come from the linked public source. Row labels identify the test set or checkpoint; unreported scores are shown as -.
| Metric rank | Model | Params | test-en WER↓ / SIM↑ | test-zh WER↓ / SIM↑ | test-zh-hard WER↓ / SIM↑ | Avg WER↓ / SIM↑ |
|---|---|---|---|---|---|---|
| - | VoxCPM 2Public weights · Inference code | 2B | 1.84 / 75.3 | 0.97 / 79.5 | 8.13 / 75.3 | 3.65 / 76.7 |
| - | dots.tts (Pretrain) | 2B | 1.34 / 76.8 | 0.96 / 80.5 | 6.46 / 79.2 | 2.92 / 78.8 |
| - | dots.tts (SOAR) | 2B | 1.30 / 77.1 | 0.94 / 81.0 | 6.60 / 79.5 | 2.95 / 79.2 |
| - | dots.tts (MF, NFE=4) | 2B | 1.29 / 76.2 | 0.94 / 80.0 | 6.60 / 78.5 | 2.94 / 78.2 |
| - | dots.tts (MF-2steps) | 2B | 1.64 / 76.4 | 1.00 / 80.4 | 6.43 / 78.5 | 3.02 / 78.4 |
| - | dots.tts (MF-2steps-STTS) | 2B | 1.41 / 75.3 | 1.04 / 79.2 | 7.83 / 76.9 | 3.43 / 77.1 |
Datasets, versions and conditions remain separate, including sources with no eligible ranking rows.
Deployable entries 6 / source rows 17
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
CosyVoice 3 · DiTAR · F5-TTS · FireRedTTS-2 · IndexTTS 2 · MegaTTS 3 · MiniMax-Speech · Qwen3-TTS · Seed-TTS · VibeVoice · dots.tts (MF-1step)
Deployable entries 4 / source rows 7
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
MiniMax · ElevenLabs · Fish-Audio S2
Deployable entries 4 / source rows 7
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
CosyVoice 2 · CosyVoice 3 (1.5B) · Fish-Audio S2
Deployable entries 4 / source rows 15
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
Gemini-2.5-Flash-TTS* · Zephyr · Gemini-2.5-Pro-TTS* · Zephyr · gpt-4o-audio-preview* · Ballad · gpt-4o-mini-tts* · Alloy · baseline: gpt-4o-mini-tts · Alloy · Qwen3-TTS · basic_ref_en · HumeAI* · — · Qwen3-TTS · Ryan · MiniMax/speech-02-hd* · EN-narr · 11Labs Multilingual v2* · Brian · F5-TTS · basic_ref_en
Deployable entries 2 / source rows 25
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
GT · Seed-DiT · MaskGCT · E2 TTS · F5 TTS · F5R-TTS · ZipVoice · Seed-ICL · SparkTTS · FireRedTTS · Qwen2.5-Omni · Qwen2.5-Omni_RL · CosyVoice · CosyVoice2 · FireRedTTS-1S · CosyVoice3-1.5B · IndexTTS2 · DiTAR · MiniMax-Speech · VoxCPM · MOSS-TTS · Qwen3-TTS · CosyVoice3.5
Scores apply to the listed dataset, metric and model version; results from different tests are not directly comparable.
The table includes only models with public weights or configurations that can run locally.
Choose a dataset to view all metrics for each model. Rows follow the official rank or selected metric. Bold marks each column’s best comparable value, including ties. Different evaluation protocols are ranked separately; missing scores are shown as -.
Deployable entries 6 / source rows 17
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
CosyVoice 3 · DiTAR · F5-TTS · FireRedTTS-2 · IndexTTS 2 · MegaTTS 3 · MiniMax-Speech · Qwen3-TTS · Seed-TTS · VibeVoice · dots.tts (MF-1step)
Values come from the linked public source. Row labels identify the test set or checkpoint; unreported scores are shown as -.
| Metric rank | Model | Params | test-en WER↓ / SIM↑ | test-zh WER↓ / SIM↑ | test-zh-hard WER↓ / SIM↑ | Avg WER↓ / SIM↑ |
|---|---|---|---|---|---|---|
| - | VoxCPM 2Public weights · Inference code | 2B | 1.84 / 75.3 | 0.97 / 79.5 | 8.13 / 75.3 | 3.65 / 76.7 |
| - | dots.tts (Pretrain) | 2B | 1.34 / 76.8 | 0.96 / 80.5 | 6.46 / 79.2 | 2.92 / 78.8 |
| - | dots.tts (SOAR) | 2B | 1.30 / 77.1 | 0.94 / 81.0 | 6.60 / 79.5 | 2.95 / 79.2 |
| - | dots.tts (MF, NFE=4) | 2B | 1.29 / 76.2 | 0.94 / 80.0 | 6.60 / 78.5 | 2.94 / 78.2 |
| - | dots.tts (MF-2steps) | 2B | 1.64 / 76.4 | 1.00 / 80.4 | 6.43 / 78.5 | 3.02 / 78.4 |
| - | dots.tts (MF-2steps-STTS) | 2B | 1.41 / 75.3 | 1.04 / 79.2 | 7.83 / 76.9 | 3.43 / 77.1 |
The table includes only models with public weights or configurations that can run locally.
Choose a dataset to view all metrics for each model. Rows follow the official rank or selected metric. Bold marks each column’s best comparable value, including ties. Different evaluation protocols are ranked separately; missing scores are shown as -.
Deployable entries 4 / source rows 7
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
MiniMax · ElevenLabs · Fish-Audio S2
Values come from the linked public source. Row labels identify the test set or checkpoint; unreported scores are shown as -.
| Metric rank | Model | Arabic↓ | Cantonese*↓ | Chinese↓ | Czech↓ | Dutch↓ | English↓ | Finnish↓ | French↓ | German↓ | Greek↓ | Hindi↓ | Indonesian↓ | Italian↓ | Japanese↓ | Korean↓ | Polish↓ | Portuguese↓ | Romanian↓ | Russian↓ | Spanish↓ | Thai↓ | Turkish↓ | Ukrainian↓ | Vietnamese↓ | Average↓ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| - | VoxCPM 2Public weights · Inference code | 13.05 / 79.1 | 38.58 / 83.5 | 1.14 / 82.5 | 24.13 / 78.3 | 0.91 / 80.8 | 2.29 / 85.4 | 2.63 / 89.0 | 4.53 / 73.5 | 0.68 / 80.3 | 2.84 / 86.0 | 19.70 / 85.6 | 1.08 / 80.0 | 1.56 / 78.0 | 4.63 / 82.8 | 1.96 / 83.3 | 1.14 / 88.4 | 1.94 / 83.7 | 21.58 / 79.7 | 3.63 / 81.1 | 1.44 / 83.1 | 2.96 / 84.0 | 0.82 / 87.1 | 6.32 / 79.8 | 3.31 / 80.6 | 5.7 / 82.3 |
| - | dots.tts (Pre.) | 37.91 / 77.5 | 37.91 / 84.7 | 1.08 / 82.3 | 5.05 / 83.8 | 1.20 / 81.4 | 1.06 / 86.9 | 3.44 / 88.0 | 3.82 / 78.2 | 1.03 / 79.5 | 2.97 / 87.6 | 14.32 / 84.5 | 2.71 / 80.8 | 3.16 / 84.5 | 7.16 / 83.1 | 5.30 / 84.3 | 2.72 / 87.3 | 1.64 / 83.1 | 3.36 / 86.2 | 3.64 / 83.0 | 0.96 / 83.9 | 7.45 / 83.8 | 5.45 / 87.4 | 1.61 / 80.5 | 3.85 / 80.7 | 6.6 / 83.5 |
| - | dots.tts (SOAR) | 36.19 / 79.1 | 42.32 / 85.0 | 0.77 / 82.5 | 4.25 / 84.2 | 1.39 / 82.2 | 1.03 / 87.5 | 4.08 / 88.3 | 3.56 / 78.6 | 1.70 / 80.6 | 3.00 / 87.6 | 14.24 / 84.7 | 2.96 / 80.8 | 3.12 / 84.7 | 5.28 / 83.7 | 5.66 / 83.6 | 3.59 / 87.8 | 2.00 / 84.3 | 3.87 / 87.1 | 4.28 / 83.2 | 1.27 / 84.0 | 7.86 / 83.9 | 4.96 / 87.3 | 1.27 / 81.2 | 3.89 / 81.6 | 6.8 / 83.9 |
| - | dots.tts (MF$_4$) | 39.65 / 77.6 | 37.82 / 84.0 | 1.01 / 81.8 | 5.67 / 83.9 | 1.30 / 82.1 | 1.09 / 86.9 | 3.61 / 88.3 | 3.26 / 78.5 | 0.91 / 79.5 | 3.19 / 87.3 | 14.75 / 84.8 | 3.91 / 81.2 | 2.16 / 84.3 | 5.17 / 83.1 | 3.93 / 84.9 | 3.42 / 87.5 | 2.40 / 83.1 | 3.38 / 86.1 | 4.42 / 83.2 | 0.80 / 84.0 | 8.03 / 84.2 | 6.20 / 86.8 | 1.66 / 80.0 | 5.43 / 80.5 | 6.8 / 83.5 |
The table includes only models with public weights or configurations that can run locally.
Choose a dataset to view all metrics for each model. Rows follow the official rank or selected metric. Bold marks each column’s best comparable value, including ties. Different evaluation protocols are ranked separately; missing scores are shown as -.
Deployable entries 4 / source rows 7
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
CosyVoice 2 · CosyVoice 3 (1.5B) · Fish-Audio S2
Values come from the linked public source. Row labels identify the test set or checkpoint; unreported scores are shown as -.
| Metric rank | Model | zh W↓ | en W↓ | hard-zh W↓ | hard-en W↓ | en→zh W↓ / S↑ | zh→en W↓ / S↑ |
|---|---|---|---|---|---|---|---|
| 1 | dots.tts (Pretrain) | 3.51 | 5.24 | 9.69 | 5.99 | 10.88 / 74.6 | 4.97 / 71.9 |
| 2 | VoxCPM 2Public weights · Inference code | 3.65 | 5.00 | 8.55 | 8.48 | - | - |
| 3 | dots.tts (SOAR) | 3.71 | 4.50 | 9.22 | 4.49 | 10.75 / 75.0 | 5.66 / 72.8 |
| 4 | dots.tts (MF, NFE=4) | 3.95 | 4.05 | 9.10 | 4.37 | 10.73 / 73.8 | 5.24 / 70.9 |
The table includes only models with public weights or configurations that can run locally.
Choose a dataset to view all metrics for each model. Rows follow the official rank or selected metric. Bold marks each column’s best comparable value, including ties. Different evaluation protocols are ranked separately; missing scores are shown as -.
Deployable entries 4 / source rows 15
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
Gemini-2.5-Flash-TTS* · Zephyr · Gemini-2.5-Pro-TTS* · Zephyr · gpt-4o-audio-preview* · Ballad · gpt-4o-mini-tts* · Alloy · baseline: gpt-4o-mini-tts · Alloy · Qwen3-TTS · basic_ref_en · HumeAI* · — · Qwen3-TTS · Ryan · MiniMax/speech-02-hd* · EN-narr · 11Labs Multilingual v2* · Brian · F5-TTS · basic_ref_en
Values come from the linked public source. Row labels identify the test set or checkpoint; unreported scores are shown as -.
| Metric rank | Model | Test setting | WER↓ | Overall↑ | Emotions↑ | Paraling.↑ | Foreign↑ | C. Pron.↑ | Quest.↑ | Syntax↑ |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | dots.tts (Pretrain) | basic_ref_en | 10.86 | 49.2% | 72.7% | 54.7% | 39.5% | 18.0% | 48.4% | 58.4% |
| 2 | dots.tts (MF4) | basic_ref_en | 11.75 | 47.9% | 59.8% | 55.2% | 36.3% | 16.7% | 50.5% | 64.8% |
| 3 | dots.tts (SOAR) | basic_ref_en | 10.45 | 47.6% | 63.9% | 52.7% | 39.4% | 16.4% | 47.0% | 65.7% |
| 4 | VoxCPM 2Public weights · Inference code | basic_ref_en | 11.84 | 41.1% | 42.3% | 44.1% | 33.3% | 18.6% | 53.4% | 52.3% |
The table includes only models with public weights or configurations that can run locally.
Choose a dataset to view all metrics for each model. Rows follow the official rank or selected metric. Bold marks each column’s best comparable value, including ties. Different evaluation protocols are ranked separately; missing scores are shown as -.
Deployable entries 2 / source rows 25
A complete public inference package for these experiments has not been verified. Other weights with the same model name do not qualify.
GT · Seed-DiT · MaskGCT · E2 TTS · F5 TTS · F5R-TTS · ZipVoice · Seed-ICL · SparkTTS · FireRedTTS · Qwen2.5-Omni · Qwen2.5-Omni_RL · CosyVoice · CosyVoice2 · FireRedTTS-1S · CosyVoice3-1.5B · IndexTTS2 · DiTAR · MiniMax-Speech · VoxCPM · MOSS-TTS · Qwen3-TTS · CosyVoice3.5
Values come from the linked public source. Row labels identify the test set or checkpoint; unreported scores are shown as -.
| Metric rank | Model | ZH CER (%) ↓ | ZH SIM ↑ | EN WER (%) ↓ | EN SIM ↑ | ZH-Hard CER (%) ↓ | ZH-Hard SIM ↑ |
|---|---|---|---|---|---|---|---|
| 1 | LongCat-AudioDiT-3.5B | 1.09 | 0.818 | 1.50 | 0.786 | 6.04 | 0.797 |
| 2 | LongCat-AudioDiT-1B | 1.18 | 0.812 | 1.78 | 0.762 | 6.33 | 0.787 |
Find processing methods and steps suited to your material.