Inflect v3 Preview
Collection
Six English voices. Four sizes. Automatic access. Non-commercial preview, not final v3. • 4 items • Updated • 1
Access the Inflect v3 Preview
Non-commercial evaluation preview. Approval is automatic. By accessing the covered materials, you agree to the Inflect Preview Evaluation License 1.0; no redistribution or hosted services. Upstream and earlier grants remain intact.
Log in or Sign Up to review the conditions and access this model content.
Inflect Nano v3 Preview
Six voices in a compact neural checkpoint.
Six English voices. Local inference. 24 kHz speech.
Preview release, not the final Inflect v3 launch. This page describes the checkpoint currently downloadable here. Sign in and accept the preview terms to request access; approval is automatic. New release-candidate components are evaluated separately, never silently substituted into this download.
Non-commercial evaluation only. The existing preview license does not permit model redistribution or hosted services. Final-release licensing remains undecided; earlier valid grants and upstream terms remain intact.
4,465,702 synthesis-core parameters · 18.93 MB FP32 checkpoint
Six fixed voices · 24 kHz mono · Offline after setup
A 4.47M-parameter neural preview with the same six voice IDs as the rest of the family. Nano remains available for evaluation and comparison; it is not selected for the current private English release candidate.
Listen first · Run locally · Explore the evidence · Release status
Six voices
Start with Cassidy and Mark, then compare all six. Each voice reads a different short passage; its passage stays the same across the four preview downloads.
CONVERSATION
Cassidy
Could you bring the smaller suitcase downstairs?
INSTRUCTIONS
Mark
Press the side button once, then wait until the indicator turns green.
EXPLANATION
Hope
A shadow becomes longer when the light source moves closer to the horizon.
NARRATION
James
Beyond the orchard, a narrow footpath curved toward the ruined tower.
DIALOGUE
Sophia
Wait! I left something important on the counter.
DIFFICULT-WORD STRESS TEST
Jasper
Photosynthesis, crystallography, and electromagnetic interference demand precise measurements.
Raw 24 kHz samples, without denoising, EQ or gain matching. Exact audio identities and downloads: sample manifest.
Choose your size
All four downloads provide the same six fixed voice IDs. Choose by footprint, then listen: a smaller model is not automatically the right trade-off for your text.
| Preview | Synthesis core | FP32 checkpoint | Position in the family |
|---|---|---|---|
| Mini | 34,310,790 | 138.39 MB | Quality-focused |
| Micro | 12,632,966 | 51.62 MB | Original Micro preview |
| Nano | 4,465,702 | 18.93 MB | Retained compact preview |
| Atom | 157,190 | 0.66 MB | Smallest synthesis core |
These are synthesis-core counts, not total frontend-inclusive parameters, install size or peak memory. Complete accounting. Nano remains available; the separate private English release-candidate selection is Mini, Fast Micro and Atom. Private candidates are not silently substituted into these downloads. Release status.
Run locally
Accept access on the model page first, then use your own Hugging Face account. In an isolated Python 3.12 environment:
python -m pip install "huggingface_hub>=0.34"
hf auth login
hf download owensong/Inflect-Nano-v3-Preview --local-dir inflect-nano-v3-preview
cd inflect-nano-v3-preview
python -m pip install -r requirements.txt "loguru==0.7.3" "attrs==26.1.0"
python -m spacy download en_core_web_sm
python synthesize.py --voice cassidy --text "What would you like to hear next?" --output example.wav
Install the expected en_core_web_sm 3.8.0 assets before offline use. The supplied
CLI, dependency prerequisites, Atom's optional CPU-LUT setup, platform caveats and
revision-pinned commands are documented in the deployment guide.
This is an evaluation package, not a certified all-platform installer.
Evaluation
Naturalness, pronunciation and runtime are different questions. We report them separately, identify the checkpoint behind each result, and retain the counts behind the percentages.
Sound quality / UTMOS v2
Higher is better. These are predicted naturalness scores, not human MOS or a listening vote. The historical Native96 panel uses 96 ordinary texts and two voice slots (192 clips). Atom was qualified later on that panel, rather than in the original joint run.
| Public checkpoint | UTMOS v2 | Scored clips |
|---|---|---|
| Inflect Mini | 3.978 | 192 |
| Inflect Micro | 3.725 | 192 |
| Inflect Nano | 3.454 | 192 |
| Inflect Atom | 2.733 | 192 |
Visual comparison, competitors and predictor limitations
The original dot plot uses an explicitly cropped 2.0-4.5 axis. Its original seven-system panel does not include the later Atom qualification. Full comparison and 0-5 view. UTMOS v2 and UTMOS22Strong are different predictors and are never pooled here.
Word accuracy / averaged WER
Lower is better. The Native96 headline is the arithmetic mean of four normalized, word-weighted corpus WERs, calculated before rounding. It is not a recognizer ensemble or a human-verified error rate. Small gaps do not establish a better-sounding model.
| Public checkpoint | ASR-mean WER | Paired clips |
|---|---|---|
| Inflect Mini | 0.269% | 192 |
| Inflect Micro | 0.185% | 192 |
| Inflect Nano | 0.926% | 192 |
| Inflect Atom | 0.176% | 192 |
Raw errors, recognizers, categories and larger test sets
WER is (substitutions + deletions + insertions) / reference words. Native96 has
2,700 reference words per recognizer. Recognizer edit counts are not counts of unique
words pronounced incorrectly. Formatting normalization does not erase missing endings
or changed meaning. Individual counts and category results.
Separate, later diagnostic runs used 500 CV3 texts and a random common cohort of 256 Seed-TTS texts. Only the published Mini and Atom identities were covered:
| Published model | CV3 / errors per reference words | Seed-256 / errors per reference words |
|---|---|---|
| Inflect Mini | 88/7377 (1.193%) | 19/2808 (0.677%) |
| Inflect Atom | 100/7377 (1.356%) | 24/2808 (0.855%) |
These later results use a substituted Qwen ASR implementation; Seed is a subset. They are not official Open TTS Leaderboard scores, not the four-ASR mean above, and not directly rankable against the leaderboard's published numbers. Original public Micro and Nano were not measured in those later runs. Protocol, coverage and limitations.
Runtime / CPU and GPU
Warm, complete-utterance timing of the original public checkpoints. Lower RTF is faster: RTF is generation wall time divided by audio duration. These are not cold-start or first-audio latency measurements.
| Published preview | CPU median RTF | RTX 5090 median RTF |
|---|---|---|
| Inflect Mini | 0.5159 | 0.0349 |
| Inflect Micro | 0.1896 | 0.0422 |
| Inflect Nano | 0.0805 | 0.0388 |
| Inflect Atom | 0.0481 | Not in this panel |
CPU: two physical AMD EPYC 7K62 cores. Atom's CPU row uses the approximate CPU-LUT backend; its default Torch backend is a different path. The original GPU panel covered Mini, Micro and Nano, not Atom. These tables must not be read as private Fast Micro results. Hardware, trials, backend choices and later Mini/Atom measurements.
Read the evaluation report · Machine-readable metrics · Historical benchmark archive
Release readiness
The current downloads remain preview releases. Their public weights, inference files, access rules and license have not changed in this documentation refresh. The working release-candidate line is Mini, FastMicro and Atom; the original Nano preview remains available here for evaluation.
Candidate validation and optional components
A separate private candidate uses stable 24 kHz defaults. Optional Mini enhancement and Cassidy-specific pitch controls are evaluated independently; neither is implied by this repository's samples, parameter count or benchmark rows. The small local four-way diagnostic found no demonstrated overall quality-proxy gain from making polish mandatory. It is not a substitute for final six-voice listening approval. No voice-cloning or paralinguistic-tag feature is promised in these downloads.
The separate candidate passed a clean, isolated Linux/CUDA install: 34 positive CLI cases and five invalid-input rejection checks. Broader checks rendered all 576 planned clips without failures, with separate CPU/GPU timing panels and completed recognition and quality-reader scoring. Final listening approval remains a separate gate. Candidate-only evidence is not a new benchmark for this published artifact. Scope and release gates.
Earlier separate R2 runtime validation
The October 31 R2 runtime measurements do not describe this published checkpoint: Nano is not included in the R2 candidate. They are not evidence about this download. Existing published quality and WER tables are unchanged; the paired ASR diagnostic applies only to the R2 Mini runtime.
Before you build
- Fixed English voices. Voice cloning, emotion/event tags and true streaming are not included.
- Listen to your actual text. Names, numbers, ambiguous readings and consonant endings can fail; unsupported or long input can be refused. Automated metrics do not certify every utterance.
- Preview terms still apply. No commercial use, model redistribution or hosted services without separate permission. Do not infer voice or data clearance from automatic access approval.
Supported scope and provenance · Current download vs release candidates · Preview license · Third-party notices
Help finish Inflect v3.
Join the Inflect Discord ·
Report a sentence ·
Explore the collection
For useful feedback, include the voice, exact non-sensitive text, repository revision, seed and what sounded wrong. Keep credentials and private voice recordings out of reports.