whatmodel.sh / tts / Magpie TTS Multilingual 357M
Magpie TTS Multilingual 357M
↗ share
+ contribute
Magpie TTS Multilingual 357M is an open-weights text-to-speech model from NVIDIA, released July 2026. Its published size is 0.364B parameters. Licence: nvidia-open-model-license.
Catalogue identity notes
MagpieTTS Multilingual v2607, the checkpoint on the main branch of nvidia/magpie_tts_multilingual_357m (card at 1980687: "July 21, 2026: MagpieTTS v2607 was released", 12 languages, "364M parameters"). Earlier checkpoints are separate releases: v2602 (tag v2602, "March 3, 2026", 9 languages) and v2512 (January 2026, 357M parameters). This entry was previously dated 2026-02 after the v2602 label, but nothing in the catalogue is tied to v2602 (no results, no API), and NVIDIA's NeMo-Speech.cpp runtime downloads the v2607 GGUF, so the entry now describes v2607. Source: https://huggingface.co/nvidia/magpie_tts_multilingual_357m/blob/19806879b16d3f2ccf28fb112b1bcd16a3c7923e/README.md
Score within the Vapi Humanness · Bradley–Terry rating, 2026-09-25 snapshot comparison set only.
$ how it compares
The text-to-speech models ranked either side of Magpie TTS Multilingual 357M, out of 148 with a score.
| # | model | score | where |
|---|---|---|---|
| 111 |  HierSpeech++ v1.1 | 39±3 | local |
| 112 |  Trendyol-TTS | 39±3 | comparison only |
| 113 | Magpie TTS Multilingual 357M | 39±6 | local |
| 114 |  Zonos v0.1 | 39±4 | api price unknown +local |
| 115 |  Maya1 | 38±6 | local |
see all text-to-speech models · find the ones that fit your machine
$ what it's good at
other tests27 not in a task · 85% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| carun tts-bench · 56-sentence English corpus · Whisper large-v3 WER | 7.0% ✎ | █····█·····█ beats 0/2 | 85% |
| Magpie held-out Arabic (ar) CER lab | 1.6% ✎ | no other model yet | — |
| Magpie held-out Arabic (ar) speaker similarity (pred-context) lab | 80.6% ✎ | no other model yet | — |
| Magpie held-out Chinese (zh) CER lab | 3.2% ✎ | no other model yet | — |
| Magpie held-out Chinese (zh) speaker similarity (pred-context) lab | 83.3% ✎ | no other model yet | — |
| Magpie held-out English (en) CER lab | 0.4% ✎ | no other model yet | — |
| Magpie held-out English (en) speaker similarity (pred-context) lab | 82.2% ✎ | no other model yet | — |
| Magpie held-out English (en) accented CER lab | 0.3% ✎ | no other model yet | — |
19 more tests
$ where to run it
Other ways to run it
- Run it as published · needs ~0.65 GB · from huggingface.co/nvidia (official)
curl -fsSL https://github.com/NVIDIA/NeMo-Speech.cpp/raw/main/scripts/install.sh | sh && export PATH="$HOME/.local/bin:$PATH" && nemo-speech synthesize "Hello from Magpie Multilingual." --output speech.wav
Official weights and documentation: huggingface.co/nvidia ↗
$ how the score was worked out · carun/tts-bench: intelligibil… 85%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 carun/tts-bench: intelligibility benchmark of four open TTS engines (published run, 2026-09-10) | 1 | 85% |
| prior · pull toward the middle | 15% | |
| weighted average | = 39.2 |
not counted (26)
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Arabic (ar) CER [correct] | 1.6% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Arabic (ar) speaker similarity (pred-context) [correct] | 80.6% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Chinese (zh) CER [correct] | 3.2% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Chinese (zh) speaker similarity (pred-context) [correct] | 83.3% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out English (en) CER [correct] | 0.4% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out English (en) speaker similarity (pred-context) [correct] | 82.2% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out English (en) accented CER [correct] | 0.3% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out English (en) accented speaker similarity (pred-context) [correct] | 74.6% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out French (fr) CER [correct] | 1.5% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out French (fr) speaker similarity (pred-context) [correct] | 74.7% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out German (de) CER [correct] | 0.8% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out German (de) speaker similarity (pred-context) [correct] | 74.2% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Hindi (hi) CER [correct] | 1.2% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Hindi (hi) speaker similarity (pred-context) [correct] | 78.8% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Italian (it) CER [correct] | 2.2% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Italian (it) speaker similarity (pred-context) [correct] | 77.3% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Japanese (ja) CER [correct] | 1.4% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Japanese (ja) speaker similarity (pred-context) [correct] | 77.5% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Korean (ko) CER [correct] | 2.7% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Korean (ko) speaker similarity (pred-context) [correct] | 80.7% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Portuguese (pt) CER [correct] | 2.9% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Portuguese (pt) speaker similarity (pred-context) [correct] | 75.3% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Spanish (es) CER [correct] | 0.6% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Spanish (es) speaker similarity (pred-context) [correct] | 79.3% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Vietnamese (vi) CER [correct] | 0.6% | fewer than three models have this test yet, or this source ran it for this model alone |
| NVIDIA MagpieTTS v2607 — original multilingual evaluation | Magpie held-out Vietnamese (vi) speaker similarity (pred-context) [correct] | 72.5% | fewer than three models have this test yet, or this source ran it for this model alone |
$ published speed measurements · 6
Reference results for the hardware, engine and settings below. They aren't verified speeds of the downloads above or predictions for your machine.
| AMD Ryzen 9 9950X3D (16C/32T) | f32 | 0.328947 x-realtime | magpie-tts.cpp f32 · median of 3 runs; seed=1234; voice=Aria; language=en; fixed short sentence; model load excluded; CPU auto threads capped at 8 GPU f32 uses default TF32 math; q8_0 retains f32 codec. Published values are rounded. | source ↗ |
| AMD Ryzen 9 9950X3D (16C/32T) | q8_0 | 0.384615 x-realtime | magpie-tts.cpp q8_0 · median of 3 runs; seed=1234; voice=Aria; language=en; fixed short sentence; model load excluded; CPU auto threads capped at 8 GPU f32 uses default TF32 math; q8_0 retains f32 codec. Published values are rounded. | source ↗ |
| NVIDIA DGX Spark (GB10 GPU) | f32 | 6.7 x-realtime | magpie-tts.cpp f32 GPU · median of 3 runs; seed=1234; voice=Aria; language=en; fixed short sentence; model load excluded; CPU auto threads capped at 8 GPU f32 uses default TF32 math; q8_0 retains f32 codec. Published values are rounded. | source ↗ |
| NVIDIA DGX Spark (GB10 GPU) | q8_0 | 10.6 x-realtime | magpie-tts.cpp q8_0 GPU · median of 3 runs; seed=1234; voice=Aria; language=en; fixed short sentence; model load excluded; CPU auto threads capped at 8 GPU f32 uses default TF32 math; q8_0 retains f32 codec. Published values are rounded. | source ↗ |
| NVIDIA DGX Spark (Grace CPU) | f32 | 0.99 x-realtime | magpie-tts.cpp f32 CPU (Grace aarch64) · median of 3 runs; seed=1234; voice=Aria; language=en; fixed short sentence; model load excluded; CPU auto threads capped at 8 GPU f32 uses default TF32 math; q8_0 retains f32 codec. Published values are rounded. | source ↗ |
| NVIDIA DGX Spark (Grace CPU) | q8_0 | 1.09 x-realtime | magpie-tts.cpp q8_0 CPU (Grace aarch64) · median of 3 runs; seed=1234; voice=Aria; language=en; fixed short sentence; model load excluded; CPU auto threads capped at 8 GPU f32 uses default TF32 math; q8_0 retains f32 codec. Published values are rounded. | source ↗ |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json