OmniVoice

# Fix list: OmniVoice

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/omnivoice, the October 2026 research run, assessed 21 October 2026. Grade C, 56.9 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 OmniVoice: 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. Reach, 40 out of 100, up to 15 more on the total

Why it scored 40: Memory earns the full 30, since the F32 files with the audio tokenizer come to 3.27 GB, about 3.9 GB with the fifth added. Runtimes score 0, because it runs only through k2-fsa's omnivoice package and community ports, not any released version of the six named runtimes. Builds score 0, with no quantised build from k2-fsa. Terms score 0, because the licence contradicts itself, CC-BY-NC with no version in the card's text, Apache 2.0 in the earlier metadata and the code, and Boson AI's community licence on the bundled tokenizer, so the stricter reading of unclear terms is taken. The ungated repository gives download the full 10. Hosted too scores 0, with only fal serving it.

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.

## 2. Risk, 35 out of 100, up to 13 more on the total

Why it scored 35: File safety gets 15, safetensors only but loaded and prompted by the publisher's package code rather than a template. Provenance gets 15 for files in k2-fsa's own organisation with hashes for every file. Licence risk falls to 0, for the non-commercial and use limits, the tokenizer licence's user threshold, and the switch from Apache 2.0 to CC-BY-NC on 3 July 2026 with no dated note. Agent guidance gets 5 for the card's disclaimer. 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.

## 3. Documentation and upkeep, 48 out of 100, up to 5.2 more on the total

Why it scored 48: The card covers intended use, architecture and languages, and misses limitations, parameters, context length and a cut-off (8). Training data is listed by source and hours in the paper and repository but without a cut-off (5). The reference code changed on 24 August 2026 (15). k2-fsa answered 8 of 31 questions on Hugging Face, not the latest licence questions (10). It's a first release with the weights unchanged (5), and the licence metadata and template don't match the card (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, 50 out of 100, up to 5 more on the total

Why it scored 50: Read for a speech model, the tool-calling, trained tool calling and structured-output lines stand for what it takes and returns. The card and README document cloning, design attributes, duration and speed controls, inline tags and the waveform output in k2-fsa's own package (25). One mode and one voice per utterance gives the one-at-a-time band (10). The output is a fixed 24 kHz waveform with no format choice or streaming (5). The chat template scores 10 because the shipped Jinja file belongs to the backbone and the prompt is built in code. Reasoning control and context score 0, with nothing to control and 8,192-token training sequences.

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.

## 5. Capability, 86 out of 100, up to 4.9 more on the total

Why it scored 86: From 5 independent benchmark results in 2 areas, each scored as its share of the way from the benchmark's floor to the best result any model has on the same leaderboard. By area, Intelligibility and speaker similarity 90, Speed 75. Publisher-submitted results count at half weight, and numbers that appear only on the publisher's card don't count.

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

How well the model does on the benchmarks that matter to an agent. We don't average a publisher's own numbers into a grade. A result counts in full when we ran it on our reference machine, or when an independent party ran it and published the method (the Berkeley Function Calling Leaderboard, Artificial Analysis, Epoch AI's Benchmarking Hub, the SWE-bench and Terminal-Bench leaderboards, the Open ASR Leaderboard and others that run models themselves). A result the publisher submitted to such a leaderboard counts at half weight. A number that appears only on the publisher's card is shown on the listing as the publisher's claim and doesn't count.

Each result is scored as its share of the way from the benchmark's floor (chance for multiple choice, zero for a pass rate) to the best result any model, open or closed, has on the same leaderboard and setting on the day we read it. A model that matches the best model anywhere scores 100 on that benchmark. For an error rate such as word error rate, lower is better and the scale runs the other way, and where the leaderboard gives no floor the rate is read against the best model's, so twice the best model's error rate scores 50. Speed as a multiple of real time (RTFx) runs from about 1 to over 20,000, so it's read on a log scale from real time to the fastest model. On the MTEB leaderboard, where most models score between 0.5 and 0.75, the floor is the median of every model the same tab lists, and its results count at half weight because MTEB doesn't say who ran them and most are submitted with the model. An arena rating (Elo) isn't a ratio, so its floor is 400 points below the best model on that arena, the gap at which the better model wins about ten games in eleven. Where a model has results at several settings (reasoning effort, say), its best one counts, because that's the setting an agent would run.

The results group into areas, each weighted within the category.

- Tool calling and agent tasks, 35 (BFCL, τ²-bench, Terminal-Bench and agent suites like them)
- Coding, 20 (SWE-bench Verified, LiveCodeBench, Aider Polyglot)
- Reasoning and knowledge, 20 (GPQA Diamond, MMLU-Pro, Humanity's Last Exam, composite indexes that publish their components)
- Instruction following, 15 (IFEval, IFBench)
- Long context, 10 (RULER, LongBench and long-context retrieval tests)

A speech model is read on its own areas, recognition accuracy (word error rate on the Open ASR Leaderboard's sets, 60) and speed (real-time factor, 40) for speech to text, and intelligibility and speaker similarity (Seed-TTS-eval, the Open TTS Leaderboard, 70) and speed (30) for text to speech. A publisher's own converted copy of the same weights (Qwen's -hf builds, say) counts as the same checkpoint. An embedding model is read on retrieval (60) and the other MTEB task types, classification, clustering, reranking and similarity among them (40), from the MTEB leaderboard. A vision model adds an image area (MMMU, ChartQA, DocVQA or OCRBench) at 20, and the others shrink to fit.

Capability is the weighted mean of the areas with results, and needs results in at least two areas (one for a speech or embedding model). With less than that it's pending for that listing, its weight is shared across the other four, and the total is capped at 61, so a model nobody has benchmarked independently can't reach B, and every model with results ranked B or better stays above it. The listing says so. Until 21 October 2026 the cap was 69, which put models nobody had benchmarked above ones that had been.

## 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.

- For the founder, the licence. The card's text says CC-BY-NC with no version and no licence file, while the code, pyproject and GitHub LICENSE say Apache 2.0 and the metadata said apache-2.0 until 3 July 2026. Whether copies taken before July stay under Apache 2.0 is a legal question we don't answer; the listing reads the licence as it is today
- Unchecked, GitHub issues on k2-fsa/OmniVoice, the channel the card names, since the GitHub API isn't enabled for this session; responsiveness rests on Hugging Face
- Training data scored 5 by the settled reading that a description without a cut-off can't go higher; a reader who takes the published source list and hours as enough would give 20
- File safety takes 10 off because the omnivoice package builds the prompt in place of a template, as the first batch read Ming-3.0-tts; a reader who counts only trust_remote_code would give 25
- Unchecked, whether a second inference provider serves OmniVoice; fal is the only one found
- Whether the fal endpoint's commercial hosting fits the non-commercial licence isn't ours to judge, and it isn't scored
- Terms of use read the self-contradicting licence strictly as unclear (0). Read as the card's CC-BY-NC alone, terms are 5 and Reach 45

## Weaknesses

- Weights are non-commercial by the card, and the bundled tokenizer adds Boson AI's licence with a 100,000-user threshold
- The licence changed from Apache 2.0 on 3 July 2026 with no dated note, and recent licence questions have no reply
- Not in any released version of Transformers, vLLM, SGLang, llama.cpp, Ollama or LM Studio, and no quantised build from k2-fsa
- No streaming, no HTTP server and no timestamps
- No safety results or security contact, and voice design is trained on Chinese and English only

## 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.

- Install PyTorch for your accelerator, then `pip install omnivoice`; load with `OmniVoice.from_pretrained('k2-fsa/OmniVoice', device_map='cuda:0', dtype=torch.float16)` (or mps, or xpu)
- Clone with `model.generate(text=..., ref_audio='ref.wav', ref_text='...')`; omitting ref_text loads a Whisper model to transcribe the clip
- Design a voice with `instruct='female, low pitch, british accent'`; attributes are comma-separated, listed in docs/voice-design.md
- Output is a list of numpy arrays at 24 kHz; write with `sf.write('out.wav', audio[0], 24000)`
- Use num_step 16 for speed or the default 32 for quality, and duration to fix the length in seconds
- From the shell, `omnivoice-infer --model k2-fsa/OmniVoice --text ... --output out.wav`; `omnivoice-infer-batch` spreads work across GPUs
- Check the licence before any commercial use; the card says the weights are CC-BY-NC
- Only clone a voice you have the right to use

## What the review panel asked for

- date the licence change
- add a versioned licence file
- a versioned licence file for the weights
- a streaming mode
- a dated licence note
- watermarking or consent checks

## 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.