Inflect-Atom-v3

language:

en pipeline_tag: text-to-speech model_name: Inflect Atom v3 license: apache-2.0 inference: false tags:
inflect
text-to-speech
english
multi-speaker
local-inference thumbnail: https://huggingface.co/owensong/Inflect-Mini-v3/resolve/7b6c0cb1ce98d8330dc96cc2685bdb71b1016837/assets/inflect-v3-hero.svg

Speech with a different shape

Atom pairs a 157,190-parameter synthesis core with DSP waveform generation, rather than a separate learned neural vocoder. The included CPU runtime takes English text and a preset voice, then writes native 24 kHz audio.

Inside this download	Specification
Learned synthesis core	157,190 parameters
Core weights	`export.pt`
Serialized core weight file	656,498 bytes (0.66 MB, decimal)
Output	Native 24 kHz, mono float WAV
Voices	Six fixed English presets
Integrity inventory	Model manifest

157,190 is the synthesis-core count, not the whole text-to-speech pipeline. The default Misaki frontend reports an additional 1,570,514 active frontend parameters. Pronunciation assets, fixed conditioning, installed dependencies, download size and runtime memory are separate quantities. The core weight file alone is not a complete installation.

Run Atom locally: this model repository is the complete v3 inference package, not just weights. Download and first generation | Choose a model and find the right files. The GitHub inflect/ folder is the older v1/v2 library, not the v3 runtime.

Voices

Six English voices. Native 24 kHz Atom output. Read each exact input below, or try your own text.

These are actual generated samples, not reference recordings. The same voice reads the same text across the v3 cards for model comparison. Different voices read different texts; this is not a voice-ranking test. Audio provenance.

Voice IDs: cassidy, mark, hope, james, sophia, jasper. No reference recording is needed.

Run locally

Start in a Python 3.12 environment and download the complete package:

python -m pip install huggingface-hub==0.36.2
hf download owensong/Inflect-Atom-v3 --local-dir inflect-atom-v3
cd inflect-atom-v3

Install the base dependencies and bundled pronunciation wheel, then verify and run:

python -m pip install -r requirements.txt
python -m pip install vendor/en_core_web_sm-3.8.0-py3-none-any.whl
python synthesize.py --verify
python synthesize.py --voice cassidy --text "A small idea can become something worth hearing." --output ../atom-cassidy.wav

Try another preset with a different output filename:

python synthesize.py --voice hope --text "A small idea can become something worth hearing." --output ../atom-hope.wav

The wrapper selects its own model directory: there is no --model atom flag. Defaults are the Misaki frontend, torch waveform backend and seed 7717. For a text file, use --text-file passage.txt instead of --text.

Use a new .wav path. The examples write beside the download, and the CLI never overwrites an existing output. Run python synthesize.py --help for all options. Speech generation runs locally after installation; setup can require downloads.

Features

Capability	Atom
English text to speech	Included
Six fixed voices	Included
Native 24 kHz WAV output	Included
DSP waveform generation	Included; default `torch` backend
Package-file integrity verification	`python synthesize.py --verify`

Optional cpu-fused and cpu-lut backends use the additional dependencies in requirements-cpu.txt. That file supplements the base requirements; it is not a complete installation recipe. Alternate backends or frontends can produce different output and are not interchangeable quality or timing guarantees. The bundled CLI is CPU-based; it does not expose a CUDA device switch.

Atom is a speech-only package. Event tags, the optional 48 kHz enhancer and VoiceTuner belong to Inflect Mini v3, not this runtime. Mini profile flags and bracketed directions such as [laugh] are not Atom controls.

Family

Three sizes, with a different synthesis core in each package.

Model	Focus
Inflect Mini v3	Larger speech core and its own optional components
Inflect Micro v3	Compact neural synthesis with an 11.79M-parameter core
Inflect Atom v3	A 157,190-parameter synthesis core with DSP waveform generation

Each repository includes its own runtime and dependencies. Family membership does not make model weights, flags or optional components interchangeable.

License and notices

Inflect-specific code and model weights are released under Apache-2.0. Read the license and component scope.

Upstream components retain their own licenses and notices. Keep the supplied third-party notices, Apache-2.0 text and StyleTTS2 MIT text with the components they cover. Pronunciation assets and installed dependencies retain their respective terms; the Inflect license does not replace those terms or grant voice/identity rights.

For questions, use model discussions, GitHub or Discord. For a reproducible issue, include the revision, voice, frontend, backend and a short text example. Keep credentials and private recordings out of public reports.