VieNeu-TTS v3 Turbo

# Fix list: VieNeu-TTS v3 Turbo

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/vieneu-tts-v3-turbo, the October 2026 research run, assessed 21 October 2026. Grade C, 58.5 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on VieNeu-TTS v3 Turbo: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Risk, 50 out of 100, up to 15.4 more on the total

Why it scored 50: File safety gets 15, safe file formats but run through the publisher's SDK, which builds prompts in code and loads the codec with remote code on GPU. Provenance gets 15 for files in the publisher's own account, linked both ways from GitHub, with hashes for every file. Apache 2.0 carries no licence deductions (15). Agent guidance gets 5 for the card's cloning warning. Safety evaluation and a security programme score 0.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-risk):

What could go wrong when you run it, and what the publisher has done about it. A higher score means less risk.

- 0 to 25, file safety. 25 for weights in safetensors or GGUF only, 15 when pickle files (.bin, .pt, .pth) sit beside them, 0 for pickle only. 10 off when the documented way to load it or to build its prompts runs code from the repository (`trust_remote_code`, or the publisher's scripts in place of a template).
- 0 to 15, provenance. 10 when the files come from the publisher's own organisation, the one its site or docs link to, and 5 for a checksum on every file.
- 0 to 20, safety evaluation. 20 when results are published and someone other than the publisher ran them or can rerun them, 10 for the publisher's own results, 0 for none.
- 0 to 15, guidance for agent use. 15 for documented prompt-injection guidance or an instruction hierarchy the model was trained to follow, 10 for a statement of what its trained refusals cover and the safeguards to add around it, 5 for general cautions, 0 for none.
- 0 to 15, licence risk. 15 to start, 5 off for each of these, a use policy or field-of-use limits beyond the law, terms the licensor can change or revoke, a revenue or user threshold, and a licence that changed without a dated note or whose link is broken. Not below 0.
- 0 to 10, a security programme. 5 for a way to report a flaw (security.txt, a disclosure policy or a security contact), 5 for advisories handled in public.

When our reference-machine runs start, measured robustness (prompt injection with AgentDojo, over-refusal with XSTest) joins Risk.

## 2. Reach, 65 out of 100, up to 13.5 more on the total

Why it scored 65: Memory earns the full 30, since every build is under 0.5 GB and the INT8 set with the codec needs about 0.3 GB. Runtimes score 0, because only the publisher's vieneu SDK and audio.cpp run it, not any released version of the six named runtimes. Builds get 10 for the publisher's own 8-bit GGUF and ONNX builds, one precision. Apache 2.0 with commercial use spelled out gives terms the full 15, the ungated files give download the full 10, and hosted too scores 0 with no provider found.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reach):

Whether an agent can run it, and where.

- 0 to 30, memory. The smallest build from the publisher or a runtime's own library that keeps 4-bit precision or better, at its file size plus a fifth for the runtime and a working context. 30 for 16 GB or less, 25 for 24 GB, 20 for 32 GB, 15 for 64 GB, 10 for what fits our reference machine (the GPU share of 128 GB of unified memory), 5 for 256 GB, 0 above.
- 0 to 25, runtimes. 5 for each of llama.cpp, Ollama, vLLM, SGLang, LM Studio and Hugging Face Transformers that runs the architecture in a released version, from the runtime's own code, library or documentation. A publisher's fork or an unreleased branch doesn't count. 25 at most.
- 0 to 15, builds from the publisher. 15 for quantised builds at two precisions or more (FP8, 4-bit, GGUF), 10 for one, 0 when only other people's quantisations exist.
- 0 to 15, terms of use. 15 for an OSI-approved licence with nothing added, 10 for a custom licence that lets anyone use it commercially for free below a threshold or under a use policy, 5 for research or non-commercial use only, 0 for no licence or unclear terms.
- 0 to 10, download. 10 when the files download with no account and no terms to accept, 5 for a click-through that needs an account, 0 when a person approves each request.
- 5, hosted too. The publisher's own API serves it, or at least two inference providers do, so an agent without the hardware can still call it.

When our reference-machine runs start, speed on that machine (tokens per second and time to first token at the build that fits) joins Reach.

## 3. Documentation and upkeep, 58 out of 100, up to 6.5 more on the total

Why it scored 58: The card covers intended use, limitations and languages, and misses the architecture's details, parameters, context length and a cut-off (8). Training data is a gated dataset of about 10,000 hours with an undisclosed pipeline (5). The files changed on 23 September and the code on 19 October 2026 (20). The publisher answered both questions on Hugging Face (20). The weights were replaced in place many times without notes (0), and the card lags the files and SDK (5).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-upkeep):

- 0 to 20, the model card. Intended use, limitations, architecture, parameters, context length, languages and knowledge cut-off. 20 for all seven, 3 off for each that's missing.
- 0 to 20, training data. 20 for the data published or listed with sources and sizes, 15 for a data card that names the sources or the mix, the cut-off and the filtering, 10 for a general description with a cut-off, 5 without one, 0 for none.
- 0 to 20, the last change to the files, the template or the reference code. 20 within 30 days, 15 within 90, 10 within 180, 5 within a year, 0 older.
- 0 to 20, responsiveness. The publisher's replies to the 40 most recent questions and reports in the repository's discussions or issues. 20 for most, 10 for some, 5 for a few, 0 for none.
- 0 to 10, version policy. 10 for earlier versions kept in their own repositories with a dated note when one is replaced, 5 for earlier versions kept without notes or a first release, 0 when files were replaced in place without a note.
- 0 to 10, file health. 10 when config, tokenizer, template and card agree and no report of a broken file waits without a reply, 5 when one does, 0 when the files contradict the card.

## 4. Agent interface, 60 out of 100, up to 6.2 more on the total

Why it scored 60: Read for a speech model, the tool-calling, trained tool calling and structured-output lines stand for what it takes and returns. The card documents voices, cloning, batching, streaming and an OpenAI-compatible speech endpoint (25), and how to ask for pcm or wav, chunked or as events (15). One voice per utterance gives the one-at-a-time band (10). The chat template scores 10, with the prompt built in the SDK. Reasoning control and context score 0, with nothing to control and 1,024 positions.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-interface):

What a model and a developer need to talk to it. Speech, image and other models with no chat interface read the tool-calling and structured-output lines for what the model takes and returns (voices, timestamps, streaming, typed outputs), and the note says how.

- 0 to 20, the chat template. 20 for a template in the repository (Jinja, in the tokenizer config or its own file), 10 when the prompt format is only given as code you run.
- 0 to 25, the tool-calling format. 25 when the publisher documents how tools are declared and how calls come back, with the parser to use in a named runtime, 10 when only the template or a runtime's own recipe says it.
- 0 to 20, tool calling the model was trained for. 20 with several calls in one turn, 10 for one call at a time.
- 0 to 15, structured output. 15 when JSON or JSON Schema output is documented for the model with how to ask for it, 5 when the card claims it without saying how.
- 10, reasoning control. Thinking that can be turned off or budgeted per request, with where the reasoning text comes back.
- 0 to 10, context. 10 for 128K tokens or more trained, 5 for 32K to 127K. Context reached only by scaling the position encoding counts one band down.

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- Unchecked, GitHub issues on pnnbao97/VieNeu-TTS, since the GitHub API isn't enabled for this session; responsiveness rests on two Hugging Face discussions, a small sample
- Builds scored 10, reading the GGUF Q8_0 and ONNX INT8 as one precision; a reader who counts GGUF as its own kind, as the line's examples do, would give 15
- Unchecked, the terms of the gated training dataset; the card says the preset-voice speakers consented, and the dataset's licence wasn't read
- File safety takes 10 off for prompts built in the SDK and the codec's remote code on GPU; the CPU ONNX path runs no remote code
- The 30-day download count is high for a model with 82 likes; the SDK fetches from this repository, but nothing here checks how the count is made up

## Weaknesses

- Only Vietnamese and English
- Runs only through the publisher's SDK; no released mainstream runtime, and the GGUF files are for audio.cpp
- Weights replaced in place many times without notes, and the card lags the SDK's voice list
- One individual maintainer, with no security contact or safety results
- Training data is a gated set with an undisclosed pipeline

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- On CPU, `pip install vieneu`, then `from vieneu import Vieneu; tts = Vieneu(); tts.save(tts.infer(text, voice='Mai Anh'), 'out.wav')`; `Vieneu(precision='int8')` is about twice as fast on a CPU with VNNI
- On GPU, install torch 2.8 for CUDA 12.8 and transformers 4.57.6 before vieneu; it switches to PyTorch with batching and stream scheduling
- Serve with `uv run python -m apps.openai_speech` from a clone of pnnbao97/VieNeu-TTS, then POST to http://localhost:8000/v1/audio/speech with model vieneu-v3-turbo, voice and response_format pcm or wav
- Call `list_preset_voices()` for the voice names of your installed version; pin the SDK version if the roster matters
- Clone with `ref_audio='voice.wav', denoise=True`, or enrol once with `add_voice(name, path)` and reuse the name
- Keep temperature near 0.8; the style argument is ignored on v3 Turbo
- Only clone a voice you have the right to use

## What the review panel asked for

- version the weights
- update the card's voice list
- the training dataset's licence on the card
- sync the card's voice list
- watermarking or consent checks
- a SECURITY.md

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.