heed-wakeword 0.1.0__tar.gz

heed_wakeword-0.1.0/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for describing the origin of the Work and
+      reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may accept and charge a
+      fee for, acceptance of support, warranty, indemnity, or other
+      liability obligations and/or rights consistent with this License.
+      However, in accepting such obligations, You may act only on Your
+      own behalf and on Your sole responsibility, not on behalf of any
+      other Contributor, and only if You agree to indemnify, defend,
+      and hold each Contributor harmless for any liability incurred by,
+      or claims asserted against, such Contributor by reason of your
+      accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Andrei Bulzan
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

heed_wakeword-0.1.0/NOTICE

@@ -0,0 +1,8 @@
+Heed Wake Word
+Copyright 2026 Andrei Bulzan
+
+This product is licensed under the Apache License, Version 2.0 (the "License").
+You may obtain a copy of the License in the accompanying LICENSE file or at
+http://www.apache.org/licenses/LICENSE-2.0
+
+Project home: https://github.com/AndreiBulzan/heed-wakeword

heed_wakeword-0.1.0/PKG-INFO

@@ -0,0 +1,251 @@
+Metadata-Version: 2.4
+Name: heed-wakeword
+Version: 0.1.0
+Summary: Tiny, custom, on-device wake-word detection. Train your own in seconds, deploy anywhere (Python / browser / mobile).
+Author: Andrei Bulzan
+License: Apache-2.0
+Project-URL: Homepage, https://github.com/AndreiBulzan/heed-wakeword
+Project-URL: Repository, https://github.com/AndreiBulzan/heed-wakeword
+Project-URL: Issues, https://github.com/AndreiBulzan/heed-wakeword/issues
+Keywords: wake-word,keyword-spotting,voice-activity,speech,on-device,edge-ai,tts,onnx,tflite
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Multimedia :: Sound/Audio :: Speech
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+License-File: NOTICE
+Requires-Dist: torch>=2.0
+Requires-Dist: numpy>=1.20
+Requires-Dist: scipy>=1.10
+Requires-Dist: soundfile>=0.12
+Requires-Dist: click>=8.0
+Provides-Extra: mic
+Requires-Dist: sounddevice>=0.4; extra == "mic"
+Provides-Extra: tts
+Requires-Dist: piper-tts>=1.2; extra == "tts"
+Provides-Extra: kokoro
+Requires-Dist: kokoro-onnx>=0.4; extra == "kokoro"
+Provides-Extra: ui
+Requires-Dist: flask>=3.0; extra == "ui"
+Provides-Extra: export
+Requires-Dist: onnx>=1.15; extra == "export"
+Requires-Dist: onnxruntime>=1.16; extra == "export"
+Provides-Extra: dev
+Requires-Dist: pytest>=7; extra == "dev"
+Provides-Extra: all
+Requires-Dist: sounddevice>=0.4; extra == "all"
+Requires-Dist: piper-tts>=1.2; extra == "all"
+Requires-Dist: kokoro-onnx>=0.4; extra == "all"
+Requires-Dist: flask>=3.0; extra == "all"
+Requires-Dist: onnx>=1.15; extra == "all"
+Requires-Dist: onnxruntime>=1.16; extra == "all"
+Dynamic: license-file
+
+# Heed Wake Word
+
+Train your own wake word in seconds, or grab a ready-made one, then run it
+fully on-device. No cloud, no telemetry, no per-call fees. A wake word here is a
+40 to 235 KB model that runs in Python, in the browser, and on iOS and Android.
+
+Heed is Apache-2.0 licensed, so commercial and closed-source use are fine, with no
+copyleft.
+
+Try it with no install: [train your own in Colab](https://colab.research.google.com/github/AndreiBulzan/heed-wakeword/blob/main/notebooks/heed_train_colab.ipynb), or run the static [browser demo](examples/inference_browser/).
+
+## Two ways to use it
+
+1. **Train a custom word.** Record a phrase a few times, or let TTS synthesize
+   it across hundreds of voices, then train on CPU or GPU in seconds and export.
+2. **Use a pretrained word.** The mobile demo bundles four example words (hey
+   doc, activate x, hey jarvis, hey fetch) and an open "custom" slot for a model
+   you train and push from the studio. hey doc and activate x are the solid
+   ones. hey jarvis and hey fetch are quick placeholders that show off live
+   multi-word switching. Slightly better pretrained defaults are planned.
+
+Both paths produce the same artifact. You get an ONNX or TFLite model plus a
+`wake.json` preprocessing contract, and it runs the same way on every platform.
+
+## Where Heed fits
+
+Tools like Picovoice (Porcupine), openWakeWord, and LiveKit are the established
+options today, and they are all good. Heed is for a specific gap: a fully
+permissive (Apache-2.0), train-your-own wake word that also runs client-side in
+the browser.
+
+In practice that means you train a custom word in seconds from the studio or the
+CLI, with multi-speaker TTS and a cross-speaker evaluation so the model works for
+people other than you. The result is a sub-250 KB model that runs the same way in
+Python, in the browser, and on iOS and Android, as ONNX (float32 or int8) or
+TFLite. You can self-host the studio in Docker or train in Colab with no setup at
+all, and commercial and closed-source use carry no per-call fees.
+
+## Quickstart (about a minute)
+
+```bash
+pip install "heed-wakeword[ui]"   # base plus the browser studio
+heed ui                            # opens http://127.0.0.1:7777
+```
+
+Record a few positives and negatives in the browser, press Train, then
+Live-test. A GPU is optional and gets used when present. If you prefer the
+terminal:
+
+```bash
+heed init my_phrase --phrase "hey computer"
+heed train my_phrase                                  # quick, tuned to your voice
+heed train my_phrase --tts-pos 400 --kokoro-pos 200   # cross-speaker, works for anyone
+heed export my_phrase                                 # wake.onnx, wake.int8.onnx, wake.tflite, wake.json
+```
+
+The package is `heed-wakeword` on PyPI. You import it as `heed`, and the command
+is `heed`.
+
+## What you get
+
+- **Tiny and fast.** small is about 10K params (41 KB), medium about 27K
+  (108 KB), large about 60K (235 KB). INT8 is roughly 40% of that. Model
+  inference takes 1 to 15 ms on a phone CPU.
+- **Many runtimes, every platform.** ONNX (fp32 and INT8) and TFLite, on
+  Python, the browser (onnxruntime-web), and React Native iOS and Android.
+- **A streaming preprocessor we wrote ourselves.** A causal high-pass with
+  50/60 Hz notches feeds a 25 ms Hann window, a 512-point FFT (a power of two,
+  so it stays fast in any language), a 40-bin log-mel, and CMN. It runs
+  incrementally, recomputing only the frames that new audio touched, and it
+  agrees with Python bit-for-bit in JS (CI checks this). On a phone, prep is
+  about 15 to 20 ms per 100 ms of audio, and an energy gate skips the model
+  during silence.
+- **Quality you can measure.** A cross-speaker held-out eval and a
+  cross-TTS-family eval tell you whether a model works beyond the trainer's own
+  voice, before you ship it.
+- **A permissive stack.** torch, numpy, scipy, soundfile, click, with optional
+  piper-tts, kokoro-onnx, flask, and onnxruntime, all under MIT, BSD, or
+  Apache-2.0. The models you train are yours to ship.
+
+## Deploy anywhere
+
+The model consumes log-mel features, so any runtime reproduces the same
+preprocessing chain. `wake.json` specifies it in full, and there are reference
+implementations in Python (`heed/audio.py`) and JS
+(`examples/*/preprocessing.js`) that agree bit-for-bit.
+
+| Target | How |
+|---|---|
+| Python | `onnxruntime` on CPU. See `export/README.md`. |
+| Browser | `onnxruntime-web` with `examples/inference_browser/`. Fully client-side and static-hostable on Vercel, Netlify, or GitHub Pages; ships a `vercel.json`. |
+| iOS and Android | `examples/inference_react_native/`, with ONNX fp32/INT8 and TFLite, plus live word and runtime switching. |
+| Other native (Flutter, Swift, Kotlin) | Run the ONNX or TFLite model, then port the preprocessing from the Python or JS reference (about 250 lines). |
+
+Deployment needs none of the training dependencies. A 3 MB runtime and your
+sub-250 KB model cover it.
+
+## Install
+
+```bash
+pip install heed-wakeword              # core: train and the model
+pip install "heed-wakeword[ui]"        # plus the browser studio (Flask)
+pip install "heed-wakeword[tts]"       # plus piper-tts, then: heed download-tts
+pip install "heed-wakeword[kokoro]"    # plus kokoro-onnx, then: heed download-kokoro
+pip install "heed-wakeword[export]"    # plus onnx and onnxruntime (export, verify)
+pip install "heed-wakeword[all]"       # everything
+heed doctor                            # check torch, onnxruntime, and TTS
+heed smoke                             # synthetic end-to-end self-test, no mic
+```
+
+## Self-host the studio (Docker)
+
+Run the browser studio in a container, with no local Python setup:
+
+```bash
+docker compose up      # builds the image, then serves http://127.0.0.1:7777
+```
+
+Recordings and trained models persist in `./workspace`. The image is CPU-only,
+which is fine for training a tiny model; for GPU training, run heed natively. See
+`Dockerfile`.
+
+## Recording good data
+
+This is the biggest lever on quality.
+
+- **Positives.** 8 to 30 recordings of the phrase. Vary your prosody, distance
+  from the mic, and room. Variety beats raw count.
+- **Negatives.** Distractor phrases in your own voice ("good morning", "the
+  weather is nice") make precious hard negatives. Add similar-sounding phrases
+  (for "hey doc", add "hey John") so the model learns the boundary.
+- **Cross-speaker.** Turn on TTS (`--tts-pos`, `--kokoro-pos`) to synthesize the
+  phrase across hundreds of voices, so the model is not tied to you. Confirm
+  with the cross-speaker eval before you ship.
+
+## CLI reference
+
+```
+heed ui              [--host 127.0.0.1] [--port 7777] [--workspace DIR]
+heed init            <name> --phrase "..."
+heed record          <name> --kind {positive|negative} --count N
+heed download-tts / download-kokoro
+heed train           <name> [--epochs N] [--tts-pos N] [--kokoro-pos N]
+                            [--target-fpr X] [--model-size {small|medium|large}] ...
+heed test            <name> <audio.wav>
+heed listen          <name>
+heed eval            <name> [--positive-dir P] [--negative-dir N]
+heed cross-tts-test  <name>
+heed export          <name>
+heed smoke / doctor
+```
+
+Run `heed <cmd> --help` for the full options.
+
+## Design, in one paragraph
+
+Log-mel spectrograms (40 bins, a 25 ms window, a 10 ms hop, a 512-point FFT)
+feed a small depthwise-separable 1D CNN over time, with a stride-2 stem, a few
+DS-conv blocks, a global average pool, and a linear head. Training builds a
+per-user set from a handful of real positives, signal-processing augmentation (a
+VTLP-style speaker warp, reverb, noise, gain), and optional multi-speaker TTS,
+with a speaker-prototype regularizer that discourages sensitivity to the
+trainer's own voice. The high-pass is causal and state-retaining, so the exact
+same filtering streams chunk by chunk on-device, and the STFT is computed
+incrementally so only the frames that new audio touched get recomputed. The
+threshold is calibrated to a target false-positive rate, and inference is a
+sliding window with an RMS and voice-band energy gate in front of the model.
+See `notes/` for the design rationale and a comparison with prior work.
+
+## GPU and CPU
+
+Training auto-detects CUDA and uses it when present, otherwise it runs on CPU.
+The model is small, so CPU training works fine and is only a little slower.
+Model inference is CPU-only by design, because the model is far too small for
+GPU offload to beat the data-transfer cost. The one place a GPU pays off is TTS
+synthesis during training. See the install notes for the optional
+`onnxruntime-gpu` swap.
+
+## Documentation
+
+Deeper guides for each tool live in [`docs/`](docs/): the
+[studio UI](docs/studio.md), [export and deploy](docs/export-and-deploy.md),
+[Colab](docs/colab.md), [Docker](docs/docker.md), [mobile](docs/mobile.md), and
+[browser and JS](docs/browser-and-js.md).
+
+## Roadmap
+
+Everything below works today: custom training from the studio or the CLI, on GPU
+or CPU; multi-speaker TTS augmentation and a cross-speaker evaluation; ONNX and
+TFLite export with verified numerical equivalence; inference in the browser and on
+iOS and Android, with live multi-word switching; a zero-install Colab trainer; a
+static client-side browser demo; and a Docker image for the studio.
+
+A few directions are interesting for later: a curated pack of speaker-independent
+phrases, more reference preprocessing ports, folding the preprocessing into the
+model graph so raw audio goes straight in, or embedded targets like TFLite-Micro.
+None of these are promised. This is a v0.1, and what runs today is the real scope.
+
+## License
+
+Apache-2.0. You can use Heed commercially, in closed-source products, with no
+obligation to open your own code. Keep the license and NOTICE file. The license
+includes a patent grant. Every dependency is MIT, BSD, or Apache-2.0.

heed_wakeword-0.1.0/README.md

@@ -0,0 +1,203 @@
+# Heed Wake Word
+
+Train your own wake word in seconds, or grab a ready-made one, then run it
+fully on-device. No cloud, no telemetry, no per-call fees. A wake word here is a
+40 to 235 KB model that runs in Python, in the browser, and on iOS and Android.
+
+Heed is Apache-2.0 licensed, so commercial and closed-source use are fine, with no
+copyleft.
+
+Try it with no install: [train your own in Colab](https://colab.research.google.com/github/AndreiBulzan/heed-wakeword/blob/main/notebooks/heed_train_colab.ipynb), or run the static [browser demo](examples/inference_browser/).
+
+## Two ways to use it
+
+1. **Train a custom word.** Record a phrase a few times, or let TTS synthesize
+   it across hundreds of voices, then train on CPU or GPU in seconds and export.
+2. **Use a pretrained word.** The mobile demo bundles four example words (hey
+   doc, activate x, hey jarvis, hey fetch) and an open "custom" slot for a model
+   you train and push from the studio. hey doc and activate x are the solid
+   ones. hey jarvis and hey fetch are quick placeholders that show off live
+   multi-word switching. Slightly better pretrained defaults are planned.
+
+Both paths produce the same artifact. You get an ONNX or TFLite model plus a
+`wake.json` preprocessing contract, and it runs the same way on every platform.
+
+## Where Heed fits
+
+Tools like Picovoice (Porcupine), openWakeWord, and LiveKit are the established
+options today, and they are all good. Heed is for a specific gap: a fully
+permissive (Apache-2.0), train-your-own wake word that also runs client-side in
+the browser.
+
+In practice that means you train a custom word in seconds from the studio or the
+CLI, with multi-speaker TTS and a cross-speaker evaluation so the model works for
+people other than you. The result is a sub-250 KB model that runs the same way in
+Python, in the browser, and on iOS and Android, as ONNX (float32 or int8) or
+TFLite. You can self-host the studio in Docker or train in Colab with no setup at
+all, and commercial and closed-source use carry no per-call fees.
+
+## Quickstart (about a minute)
+
+```bash
+pip install "heed-wakeword[ui]"   # base plus the browser studio
+heed ui                            # opens http://127.0.0.1:7777
+```
+
+Record a few positives and negatives in the browser, press Train, then
+Live-test. A GPU is optional and gets used when present. If you prefer the
+terminal:
+
+```bash
+heed init my_phrase --phrase "hey computer"
+heed train my_phrase                                  # quick, tuned to your voice
+heed train my_phrase --tts-pos 400 --kokoro-pos 200   # cross-speaker, works for anyone
+heed export my_phrase                                 # wake.onnx, wake.int8.onnx, wake.tflite, wake.json
+```
+
+The package is `heed-wakeword` on PyPI. You import it as `heed`, and the command
+is `heed`.
+
+## What you get
+
+- **Tiny and fast.** small is about 10K params (41 KB), medium about 27K
+  (108 KB), large about 60K (235 KB). INT8 is roughly 40% of that. Model
+  inference takes 1 to 15 ms on a phone CPU.
+- **Many runtimes, every platform.** ONNX (fp32 and INT8) and TFLite, on
+  Python, the browser (onnxruntime-web), and React Native iOS and Android.
+- **A streaming preprocessor we wrote ourselves.** A causal high-pass with
+  50/60 Hz notches feeds a 25 ms Hann window, a 512-point FFT (a power of two,
+  so it stays fast in any language), a 40-bin log-mel, and CMN. It runs
+  incrementally, recomputing only the frames that new audio touched, and it
+  agrees with Python bit-for-bit in JS (CI checks this). On a phone, prep is
+  about 15 to 20 ms per 100 ms of audio, and an energy gate skips the model
+  during silence.
+- **Quality you can measure.** A cross-speaker held-out eval and a
+  cross-TTS-family eval tell you whether a model works beyond the trainer's own
+  voice, before you ship it.
+- **A permissive stack.** torch, numpy, scipy, soundfile, click, with optional
+  piper-tts, kokoro-onnx, flask, and onnxruntime, all under MIT, BSD, or
+  Apache-2.0. The models you train are yours to ship.
+
+## Deploy anywhere
+
+The model consumes log-mel features, so any runtime reproduces the same
+preprocessing chain. `wake.json` specifies it in full, and there are reference
+implementations in Python (`heed/audio.py`) and JS
+(`examples/*/preprocessing.js`) that agree bit-for-bit.
+
+| Target | How |
+|---|---|
+| Python | `onnxruntime` on CPU. See `export/README.md`. |
+| Browser | `onnxruntime-web` with `examples/inference_browser/`. Fully client-side and static-hostable on Vercel, Netlify, or GitHub Pages; ships a `vercel.json`. |
+| iOS and Android | `examples/inference_react_native/`, with ONNX fp32/INT8 and TFLite, plus live word and runtime switching. |
+| Other native (Flutter, Swift, Kotlin) | Run the ONNX or TFLite model, then port the preprocessing from the Python or JS reference (about 250 lines). |
+
+Deployment needs none of the training dependencies. A 3 MB runtime and your
+sub-250 KB model cover it.
+
+## Install
+
+```bash
+pip install heed-wakeword              # core: train and the model
+pip install "heed-wakeword[ui]"        # plus the browser studio (Flask)
+pip install "heed-wakeword[tts]"       # plus piper-tts, then: heed download-tts
+pip install "heed-wakeword[kokoro]"    # plus kokoro-onnx, then: heed download-kokoro
+pip install "heed-wakeword[export]"    # plus onnx and onnxruntime (export, verify)
+pip install "heed-wakeword[all]"       # everything
+heed doctor                            # check torch, onnxruntime, and TTS
+heed smoke                             # synthetic end-to-end self-test, no mic
+```
+
+## Self-host the studio (Docker)
+
+Run the browser studio in a container, with no local Python setup:
+
+```bash
+docker compose up      # builds the image, then serves http://127.0.0.1:7777
+```
+
+Recordings and trained models persist in `./workspace`. The image is CPU-only,
+which is fine for training a tiny model; for GPU training, run heed natively. See
+`Dockerfile`.
+
+## Recording good data
+
+This is the biggest lever on quality.
+
+- **Positives.** 8 to 30 recordings of the phrase. Vary your prosody, distance
+  from the mic, and room. Variety beats raw count.
+- **Negatives.** Distractor phrases in your own voice ("good morning", "the
+  weather is nice") make precious hard negatives. Add similar-sounding phrases
+  (for "hey doc", add "hey John") so the model learns the boundary.
+- **Cross-speaker.** Turn on TTS (`--tts-pos`, `--kokoro-pos`) to synthesize the
+  phrase across hundreds of voices, so the model is not tied to you. Confirm
+  with the cross-speaker eval before you ship.
+
+## CLI reference
+
+```
+heed ui              [--host 127.0.0.1] [--port 7777] [--workspace DIR]
+heed init            <name> --phrase "..."
+heed record          <name> --kind {positive|negative} --count N
+heed download-tts / download-kokoro
+heed train           <name> [--epochs N] [--tts-pos N] [--kokoro-pos N]
+                            [--target-fpr X] [--model-size {small|medium|large}] ...
+heed test            <name> <audio.wav>
+heed listen          <name>
+heed eval            <name> [--positive-dir P] [--negative-dir N]
+heed cross-tts-test  <name>
+heed export          <name>
+heed smoke / doctor
+```
+
+Run `heed <cmd> --help` for the full options.
+
+## Design, in one paragraph
+
+Log-mel spectrograms (40 bins, a 25 ms window, a 10 ms hop, a 512-point FFT)
+feed a small depthwise-separable 1D CNN over time, with a stride-2 stem, a few
+DS-conv blocks, a global average pool, and a linear head. Training builds a
+per-user set from a handful of real positives, signal-processing augmentation (a
+VTLP-style speaker warp, reverb, noise, gain), and optional multi-speaker TTS,
+with a speaker-prototype regularizer that discourages sensitivity to the
+trainer's own voice. The high-pass is causal and state-retaining, so the exact
+same filtering streams chunk by chunk on-device, and the STFT is computed
+incrementally so only the frames that new audio touched get recomputed. The
+threshold is calibrated to a target false-positive rate, and inference is a
+sliding window with an RMS and voice-band energy gate in front of the model.
+See `notes/` for the design rationale and a comparison with prior work.
+
+## GPU and CPU
+
+Training auto-detects CUDA and uses it when present, otherwise it runs on CPU.
+The model is small, so CPU training works fine and is only a little slower.
+Model inference is CPU-only by design, because the model is far too small for
+GPU offload to beat the data-transfer cost. The one place a GPU pays off is TTS
+synthesis during training. See the install notes for the optional
+`onnxruntime-gpu` swap.
+
+## Documentation
+
+Deeper guides for each tool live in [`docs/`](docs/): the
+[studio UI](docs/studio.md), [export and deploy](docs/export-and-deploy.md),
+[Colab](docs/colab.md), [Docker](docs/docker.md), [mobile](docs/mobile.md), and
+[browser and JS](docs/browser-and-js.md).
+
+## Roadmap
+
+Everything below works today: custom training from the studio or the CLI, on GPU
+or CPU; multi-speaker TTS augmentation and a cross-speaker evaluation; ONNX and
+TFLite export with verified numerical equivalence; inference in the browser and on
+iOS and Android, with live multi-word switching; a zero-install Colab trainer; a
+static client-side browser demo; and a Docker image for the studio.
+
+A few directions are interesting for later: a curated pack of speaker-independent
+phrases, more reference preprocessing ports, folding the preprocessing into the
+model graph so raw audio goes straight in, or embedded targets like TFLite-Micro.
+None of these are promised. This is a v0.1, and what runs today is the real scope.
+
+## License
+
+Apache-2.0. You can use Heed commercially, in closed-source products, with no
+obligation to open your own code. Keep the license and NOTICE file. The license
+includes a patent grant. Every dependency is MIT, BSD, or Apache-2.0.