Release NVIDIA WakeWord Trainer v12

Update prebuilt firmware flasher and tagged Docker releases

.github/workflows/docker-publish.yml

@@ -0,0 +1,144 @@
+name: Publish Docker Images
+
+on:
+  push:
+    tags:
+      - "v*"
+  workflow_dispatch:
+
+permissions:
+  contents: write
+  packages: write
+
+concurrency:
+  group: docker-publish-${{ github.ref }}
+  cancel-in-progress: true
+
+env:
+  REGISTRY: ghcr.io
+  IMAGE_NAME: tatertotterson/microwakeword
+
+jobs:
+  docker:
+    name: Docker image
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v4
+
+      - name: Validate tag matches trainer version
+        if: startsWith(github.ref, 'refs/tags/')
+        shell: bash
+        run: |
+          set -euo pipefail
+          version="$(tr -d '[:space:]' < VERSION)"
+          expected_tag="v${version#v}"
+          if [[ "${GITHUB_REF_NAME}" != "${expected_tag}" ]]; then
+            echo "Tag ${GITHUB_REF_NAME} does not match trainer version ${expected_tag}." >&2
+            exit 1
+          fi
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Log in to GHCR
+        uses: docker/login-action@v3
+        with:
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Docker metadata
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
+          flavor: |
+            latest=false
+          tags: |
+            type=raw,value=latest
+            type=ref,event=tag
+
+      - name: Build and push image
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: dockerfile
+          platforms: linux/amd64
+          push: true
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+          cache-from: type=gha,scope=mww-trainer-nvidia-docker
+          cache-to: type=gha,mode=max,scope=mww-trainer-nvidia-docker
+
+      - name: Docker metadata (Blackwell)
+        id: meta-blackwell
+        uses: docker/metadata-action@v5
+        with:
+          images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
+          flavor: |
+            latest=false
+          tags: |
+            type=raw,value=blackwell
+            type=ref,event=tag,suffix=-blackwell
+
+      - name: Build and push Blackwell image
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: dockerfile.blackwell
+          platforms: linux/amd64
+          push: true
+          tags: ${{ steps.meta-blackwell.outputs.tags }}
+          labels: ${{ steps.meta-blackwell.outputs.labels }}
+          cache-from: type=gha,scope=mww-trainer-nvidia-docker-blackwell
+          cache-to: type=gha,mode=max,scope=mww-trainer-nvidia-docker-blackwell
+
+      - name: Create release notes
+        if: startsWith(github.ref, 'refs/tags/')
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          TAG_NAME: ${{ github.ref_name }}
+          REPO: ${{ github.repository }}
+        run: |
+          set -euo pipefail
+
+          title="microWakeWord Nvidia Trainer ${TAG_NAME}"
+          generated_notes="$(mktemp)"
+          release_notes="$(mktemp)"
+          test -s WHATS_NEW.md
+
+          gh api "repos/${REPO}/releases/generate-notes" \
+            -f tag_name="${TAG_NAME}" \
+            -f target_commitish="${GITHUB_SHA}" \
+            --jq '.body' > "${generated_notes}"
+
+          {
+            echo "## What's New"
+            echo
+            cat WHATS_NEW.md
+            echo
+            echo
+            echo "## Docker Images"
+            echo
+            echo "- \`ghcr.io/tatertotterson/microwakeword:${TAG_NAME}\`"
+            echo "- \`ghcr.io/tatertotterson/microwakeword:latest\`"
+            echo "- \`ghcr.io/tatertotterson/microwakeword:${TAG_NAME}-blackwell\`"
+            echo "- \`ghcr.io/tatertotterson/microwakeword:blackwell\`"
+            echo
+            cat "${generated_notes}"
+          } > "${release_notes}"
+
+          if gh release view "${TAG_NAME}" >/dev/null 2>&1; then
+            gh release edit "${TAG_NAME}" \
+              --title "${title}" \
+              --notes-file "${release_notes}" \
+              --latest \
+              --verify-tag
+          else
+            gh release create "${TAG_NAME}" \
+              --title "${title}" \
+              --notes-file "${release_notes}" \
+              --latest \
+              --verify-tag
+          fi

.gitignore

@@ -1,3 +1,4 @@
 personal_samples/*
 data/
+trim_history/
 .DS_Store

README.md

@@ -1,16 +1,15 @@
 <div align="center">
-  <h1>microWakeWord NVIDIA Docker Trainer UI</h1>
-  <img width="800" alt="Screenshot 2026-04-14 at 11 02 06 PM" src="https://github.com/user-attachments/assets/694f4cb7-e4d8-4e2b-80ec-b40fb41cbfff" />
+  <a href="https://taterassistant.com">
+    <img src="images/tater-repo-logo.png" alt="microWakeWord Trainer" width="460"/>
+  </a>
 </div>
+<h3 align="center">
+  <a href="https://taterassistant.com">taterassistant.com</a>
+</h3>
 
-Train custom microWakeWord models in Docker with:
+Train custom microWakeWord models in Docker with NVIDIA/CUDA acceleration, generated Piper samples, device-captured samples, reviewed false-wake negatives, live training logs, and local wake-word links for Tater Native satellites.
 
-- uploaded personal voice samples
-- automatically generated Piper TTS samples
-- a browser-based trainer UI
-- live training logs in a popup console
-
-This project no longer records audio in the browser. The UI is now upload-first: users add their own audio files, the app validates or converts them, and training runs from the same page.
+Real samples come from device-captured wake audio, close misses, or manual uploads. Every saved sample is normalized to `16 kHz / mono / 16-bit PCM WAV` before training.
 
 ---
 
@@ -20,6 +19,27 @@ This project no longer records audio in the browser. The UI is now upload-first:
 docker pull ghcr.io/tatertotterson/microwakeword:latest
 ```
 
+Tagged releases also publish matching immutable image tags:
+
+```bash
+docker pull ghcr.io/tatertotterson/microwakeword:v12
+```
+
+The release tag must match `VERSION`. Update `WHATS_NEW.md` before tagging; the Docker workflow prepends it to GitHub's automatically generated release notes.
+
+RTX 50-series / Blackwell GPUs use a separate image with CUDA 12.8 and a
+Python 3.13 TensorFlow build for `sm_120`:
+
+```bash
+docker pull ghcr.io/tatertotterson/microwakeword:blackwell
+docker pull ghcr.io/tatertotterson/microwakeword:v12-blackwell
+```
+
+Use the Blackwell image only for RTX 50-series cards. It includes the
+community-built TensorFlow wheel from
+[chivitiH/tensorflow-blackwell-python313](https://github.com/chivitiH/tensorflow-blackwell-python313),
+which is unofficial and licensed CC BY-NC 4.0.
+
 ---
 
 ## Run The Container
@@ -27,41 +47,104 @@ docker pull ghcr.io/tatertotterson/microwakeword:latest
 ```bash
 docker run -d \
   --gpus all \
-  -p 8888:8888 \
+  --network host \
+  -e REC_PORT=8789 \
   -v $(pwd):/data \
   ghcr.io/tatertotterson/microwakeword:latest
 ```
 
-What these flags do:
+Use a version tag such as `ghcr.io/tatertotterson/microwakeword:v12` when you want to pin a known release instead of tracking `latest`.
+For RTX 50-series cards, use `ghcr.io/tatertotterson/microwakeword:blackwell`
+or a pinned tag such as `ghcr.io/tatertotterson/microwakeword:v12-blackwell`
+in the same `docker run` command.
 
-- `--gpus all` enables GPU acceleration
-- `-p 8888:8888` exposes the trainer UI
-- `-v $(pwd):/data` persists models, downloaded voices, datasets, and personal samples
+The flags:
 
-Then open:
+- `--gpus all` enables GPU acceleration.
+- `--network host` exposes the trainer server directly so satellites can send captured audio and load trained wake-word files.
+- `-e REC_PORT=8789` sets the trainer web UI and captured-audio port. Change this value if `8789` is already in use.
+- `-v $(pwd):/data` persists models, downloaded voices, datasets, samples, and generated wake-word artifacts.
+
+If you do not use host networking, publish the trainer port and make sure satellites can reach it from your LAN.
+
+Open:
 
 ```text
-http://localhost:8888
+http://localhost:8789
 ```
 
+If you change `REC_PORT`, open that port instead and use the same port in the satellite `Trainer App URL`.
+
 ---
 
 ## What The UI Does
 
-- Start a wake word session
-- Test TTS pronunciation
-- Upload one or many personal samples
-- Normalize uploads to `16 kHz / mono / 16-bit PCM WAV`
-- Train with or without personal samples
-- Show a popup console with live progress and logs
-
-Personal samples are optional. If none are uploaded, the trainer can still proceed with TTS-only data after confirmation.
+- `Trainer` starts a wake-word session, shows positive/negative sample counts, and launches training.
+- `Auto Training` transcribes real wake triggers, promotes phrase-misses to hard negatives, schedules retraining, and refreshes Tater Native satellites.
+- `Captured Audio` reviews clips sent by Tater Native or ESPHome sats, including wake hits, close misses, and false wakes.
+- `Samples` plays, removes, clears, and manually imports personal or negative samples.
+- `Wake Words` lists locally trained JSON/model links for live wake-word switching in Tater.
+- Popup consoles show colorized training logs while long-running jobs are active.
 
 ---
 
-## Personal Samples
+## Captured Audio Workflow
 
-Accepted upload formats include:
+To collect samples from a sat, point its trainer feedback setting at this app. Tater Native satellites use the native settings popup in Tater. Older ESPHome satellites can still use their device entities.
+
+For Tater Native satellites, enable trainer feedback in Tater:
+
+- `Send Good Wakes To Trainer` toggles upload of confirmed wake-word triggers.
+- `Send Close Misses To Trainer` toggles upload of near misses.
+- `Trainer App URL` sets the trainer address, for example `http://trainer.local:8789` or `http://<trainer-ip>:8789`.
+
+For older ESPHome firmware, the equivalent capture setup is exposed as device entities:
+
+- `Capture Wake Audio` toggles upload of wake-word triggers.
+- `Capture Close Misses` toggles upload of near misses.
+- `Trainer App URL` sets the trainer address, for example `http://<trainer-ip>:8789`.
+
+Satellites send raw captured audio to:
+
+```text
+/api/upload_captured_audio_raw
+```
+
+Keep the training app running and reachable at the `Trainer App URL` while capture is enabled. The sats upload clips live; if the app is stopped or the URL is wrong, captured audio will not be saved.
+
+In the `Captured Audio` tab:
+
+- play each clip from the inbox
+- mark good wake-word clips as `This is good`
+- mark bad triggers as `False wake`
+- discard clips that should not be used
+
+Approved clips move into:
+
+```text
+/data/personal_samples/
+```
+
+False wakes move into:
+
+```text
+/data/negative_samples/
+```
+
+Captured audio is boosted for easier playback in the UI, then kept in the correct training format.
+
+---
+
+## Samples
+
+The `Samples` tab is the sample library.
+
+- `Personal` samples are positive examples of the wake word.
+- `Negative` samples are reviewed false wakes or hard negatives.
+- Both can be played back and removed one at a time.
+- Manual upload is available here as an optional seed path.
+
+Accepted manual upload formats include:
 
 - WAV
 - MP3
@@ -72,97 +155,178 @@ Accepted upload formats include:
 - OPUS
 - WEBM
 
-The backend validates or converts uploads with `ffmpeg` and stores the normalized files in:
+Uploads are validated or converted with `ffmpeg` into:
 
 ```text
-/data/personal_samples/
+16 kHz / mono / 16-bit PCM WAV
 ```
 
-Notes:
+Starting a new session does not clear samples. Use the clear buttons in `Samples` if you want to remove saved personal or negative clips.
 
-- starting a new session does not clear personal samples
-- use the `Clear personal samples` button if you want to wipe them
-- any uploaded personal samples are automatically included in training
+---
+
+## Auto Training
+
+`Auto Training` is an opt-in false-positive loop. It is disabled until you enter the exact wake phrase and enable it.
+
+For each new wake-trigger clip sent to the trainer:
+
+1. Faster Whisper transcribes the audio locally.
+2. If the transcript contains the configured wake phrase, the clip stays in `Captured Audio` for manual positive review.
+3. If speech was transcribed but the wake phrase is absent, the clip moves to `/data/negative_samples/` as an auto-reviewed hard negative.
+4. Empty transcripts, close misses, VAD-blocked captures, and captures for another wake word stay out of the automatic negative path.
+
+The default `small.en` model uses CUDA with `float16` when CTranslate2 can see an NVIDIA GPU, and falls back to CPU with `int8`. Choose a multilingual Faster Whisper model such as `small` when the wake phrase is not English. Downloaded STT models are cached in `/data/auto_train_models/`.
+
+Scheduled training runs only after the configured number of new automatic negatives has accumulated. A successful run publishes the replacement model at the same wake-word URL and can call Tater's native satellite settings API to make connected satellites pull it again. This refresh uses the existing Tater Native update path, so no satellite firmware change is required.
+
+The `Trainer public URL` must be reachable from the satellites. With the documented `--network host` command, the trainer can normally use the LAN address from the browser request or host network. If you open the UI as `http://localhost:8789`, enter a value such as `http://192.168.1.50:8789`, or start the container with `REC_PUBLIC_BASE_URL` set to that value. When using Docker bridge networking, always set this URL to the published host address; a container bridge address is not satellite-reachable.
+
+The default Tater URL, `http://127.0.0.1:8501`, assumes the documented host networking. Change it to a container-reachable Tater address if you use another Docker network. The optional API token is stored in `/data/auto_train_config.json` with owner-only permissions.
+
+---
+
+## Training Flow
+
+1. Enter the wake phrase in `Trainer`.
+2. Choose the language.
+3. Optionally test pronunciation with `Test TTS`.
+4. Review the positive and negative sample counts.
+5. Click `Start training`.
+6. Watch the popup training console.
+
+Personal samples are optional. Training can run with zero personal samples after confirmation, using generated TTS samples and the stock negative datasets.
+
+Reviewed negative samples are converted into `/data/work/reviewed_negative_features/` and inserted into the training YAML as a hard-negative feature set when present.
+
+On RTX 50-series / Blackwell GPUs, the Blackwell Docker image keeps sample generation and augmentation in the normal Python 3.12 trainer environment, then runs only the TensorFlow training/export stage in `/data/.venv-blackwell` with Python 3.13 and the Blackwell-native TensorFlow wheel.
 
 ---
 
 ## Language Support
 
-The language selector is dynamic.
+The language picker is dynamic.
 
-- `en` is always available
-- non-English languages are populated from Piper voice metadata
-- when you train with a non-English language, the backend downloads all Piper ONNX voices for that selected language only
-- it does not pre-download every language
-- already-downloaded voices are reused on later runs
+- `en` is always available.
+- English keeps the existing dedicated generator model path.
+- Non-English languages are discovered from the Piper voices catalog and any local Piper voice metadata.
+- When a non-English language is selected, the trainer downloads all voices for that selected language only.
+- Already-downloaded voices are reused.
+- It does not download every language up front.
 
-English stays on its existing dedicated generator model path. Non-English languages use the selected language's ONNX Piper voices.
-
-If the Piper catalog is unavailable, already-installed local voices can still be used.
+If the upstream Piper catalog is unavailable, already-installed local voices are used when available.
 
 ---
 
-## Training Behavior
+## Dataset Behavior
 
-1. Enter the wake word
-2. Optionally test pronunciation
-3. Optionally upload personal samples
-4. Click `Start training`
-5. Watch the popup console for:
-   - selected-language voice downloads when needed
-   - sample generation progress
-   - dataset setup
-   - training progress and completion
-
-The `Open console` button lets you reopen the log window after closing it.
-
----
-
-## First Run Notes
-
-The first real training run may download large training assets into `/data`, such as:
+The first training run downloads and prepares missing training assets into `/data`, including:
 
 - Piper voices for the selected language
-- training datasets and background data
-- Python training environment dependencies
+- negative datasets and background data
+- the Python training environment
+- generated samples and augmented feature caches
 
-These are reused later unless you delete `/data`.
+After those assets are prepared, later runs reuse the local copies unless the mounted `/data` contents are deleted.
+
+---
+
+## Trained Wake Words
+
+The `Wake Words` tab lists locally trained wake-word packages from `/data/trained_wake_words/`.
+
+- Copy the JSON URL into the Tater Native satellite settings to switch wake words live.
+- Links use the configured public trainer URL, a non-loopback browser host, or the detected LAN address instead of advertising `127.0.0.1` to satellites.
+- Open the JSON or model links directly for quick inspection.
+- The JSON includes the matching model path plus Tater tuning metadata.
+- No firmware flashing happens from this trainer app anymore.
+
+Use the main Tater app for satellite firmware updates and USB flashing.
 
 ---
 
 ## Output Files
 
-Successful runs produce:
+Successful runs produce timestamped training output folders such as:
 
 ```text
-/data/output/<wake_word>.tflite
-/data/output/<wake_word>.json
+/data/output/<timestamp>-<wake_word>-<samples>-<steps>/<wake_word>.tflite
+/data/output/<timestamp>-<wake_word>-<samples>-<steps>/<wake_word>.json
 ```
 
-If those files already exist, the trainer creates timestamped backups before replacing them.
+The trainer also syncs Tater-ready wake-word artifacts into:
+
+```text
+/data/trained_wake_words/<wake_word>.tflite
+/data/trained_wake_words/<wake_word>.json
+```
+
+The `Wake Words` tab uses `/data/trained_wake_words/` to populate the local wake-word links.
+
+The JSON keeps the standard microWakeWord fields for compatibility:
+
+```json
+{
+  "micro": {
+    "probability_cutoff": 0.97,
+    "sliding_window_size": 6
+  }
+}
+```
+
+It also includes Tater Native metadata used by newer satellites and the Tater settings UI:
+
+```json
+{
+  "model_format": "tflite_stream_state_internal_quant",
+  "quantization": "int8",
+  "sample_rate": 16000,
+  "tater_native": {
+    "format_version": 1,
+    "wake_threshold": 0.97,
+    "wake_sliding_window": 6,
+    "close_miss_threshold": 0.80,
+    "frontend": {
+      "name": "tflm_microfrontend",
+      "sample_rate": 16000,
+      "feature_duration_ms": 30,
+      "feature_step_ms": 10,
+      "feature_size": 40
+    }
+  }
+}
+```
+
+Calibration metrics are included under `calibration` so false accepts/hour and recall can be surfaced in the UI.
+Calibration evaluates thresholds from `0.95` through `1.00` with sliding windows of `5`, `6`, and `7`. Among candidates within 0.5 percentage points of the best recall, it prefers the lowest measured ambient false-accept rate. If calibration cannot complete, packaging uses the conservative `0.97` threshold and a window of `6`.
 
 ---
 
 ## Resetting Everything
 
-If you want a clean slate, stop the container and remove the contents of your mounted `/data` directory.
+If you want a clean slate, stop the container and remove the contents of the mounted `/data` directory.
 
-That will remove:
+That removes:
 
 - personal samples
+- negative samples
+- captured inbox clips
 - downloaded Piper voices
 - cached datasets
 - training environments
 - trained models
+- Auto Training settings, state, transcripts, and cached Faster Whisper models
 
 ---
 
-## Notes
+## Important Notes
 
-- browser microphone recording has been removed
-- personal samples are optional
-- the server module is now `trainer_server.py`
-- the launcher script is now `run.sh`
+- Personal samples are optional.
+- Negative samples are optional but useful for reducing false wakes.
+- Auto Training is disabled by default and only classifies actual wake triggers automatically.
+- The UI server is `trainer_server.py`.
+- The launcher is `run.sh`.
+- Trainer capture settings live in Tater for Tater Native satellites, and on device entities for older ESPHome satellites.
 
 ---
 
@@ -172,3 +336,4 @@ Built on top of:
 
 - [microWakeWord](https://github.com/kahrendt/microWakeWord)
 - [piper-sample-generator](https://github.com/rhasspy/piper-sample-generator)
+- [tensorflow-blackwell-python313](https://github.com/chivitiH/tensorflow-blackwell-python313) for the optional RTX 50-series / Blackwell image

VERSION

@@ -0,0 +1 @@
+12

WHATS_NEW.md

@@ -0,0 +1,5 @@
+- Added opt-in Auto Training for false-positive wake triggers, using Faster Whisper with automatic CUDA/float16 selection and CPU/int8 fallback.
+- Wake triggers whose transcripts do not contain the configured phrase can now become hard negatives automatically; close misses, empty transcripts, and phrase matches remain available for manual review.
+- Added scheduled retraining with a minimum-new-negatives threshold and automatic Tater Native satellite refresh after a successful model build.
+- Wake-word download links now advertise a LAN-reachable trainer URL instead of `127.0.0.1`.
+- Tightened detector calibration defaults to favor fewer ambient false accepts while preserving candidates within 0.5 percentage points of the best recall.

cli/calibrate_detector.py

@@ -9,24 +9,22 @@ import math
 import os
 from datetime import datetime, timezone
 from pathlib import Path
-from typing import Iterable, Sequence
+from typing import Any, Iterable, Sequence
 
 import numpy as np
 import yaml
 
-from microwakeword.data import FeatureHandler
-from microwakeword.inference import Model
-
-
-DEFAULT_WINDOW_SIZES = [3, 4, 5, 6, 7]
-DEFAULT_TARGET_FAPH = float(os.environ.get("MWW_CALIBRATION_TARGET_FAPH", "1.0"))
+DEFAULT_WINDOW_SIZES = [5, 6, 7]
+DEFAULT_TARGET_FAPH = float(os.environ.get("MWW_CALIBRATION_TARGET_FAPH", "0.25"))
 DEFAULT_COOLDOWN_SLICES = int(os.environ.get("MWW_CALIBRATION_COOLDOWN_SLICES", "25"))
 DEFAULT_POSITIVE_SKIP_SLICES = int(
     os.environ.get("MWW_CALIBRATION_POSITIVE_SKIP_SLICES", "25")
 )
 DEFAULT_CUTOFF_STEP = float(os.environ.get("MWW_CALIBRATION_CUTOFF_STEP", "0.01"))
-DEFAULT_CUTOFF_MIN = float(os.environ.get("MWW_CALIBRATION_CUTOFF_MIN", "0.00"))
+DEFAULT_CUTOFF_MIN = float(os.environ.get("MWW_CALIBRATION_CUTOFF_MIN", "0.95"))
 DEFAULT_CUTOFF_MAX = float(os.environ.get("MWW_CALIBRATION_CUTOFF_MAX", "1.00"))
+DEFAULT_RECALL_MARGIN = float(os.environ.get("MWW_CALIBRATION_RECALL_MARGIN", "0.005"))
+PREFERRED_WINDOW_SIZE = 6
 
 
 def parse_args() -> argparse.Namespace:
@@ -65,6 +63,15 @@ def parse_args() -> argparse.Namespace:
         default=DEFAULT_TARGET_FAPH,
         help="Target ambient false accepts per hour for the selected operating point.",
     )
+    parser.add_argument(
+        "--recall-margin",
+        type=float,
+        default=DEFAULT_RECALL_MARGIN,
+        help=(
+            "Maximum recall loss allowed when preferring a candidate with fewer "
+            "ambient false accepts (0.005 means 0.5 percentage points)."
+        ),
+    )
     parser.add_argument(
         "--cooldown-slices",
         type=int,
@@ -159,7 +166,13 @@ def _compute_false_accepts_per_hour(
 def _select_best_candidate(
     candidates: list[dict[str, float]],
     target_faph: float,
+    recall_margin: float = DEFAULT_RECALL_MARGIN,
 ) -> tuple[dict[str, float], float]:
+    if not candidates:
+        raise ValueError("at least one calibration candidate is required")
+    if recall_margin < 0:
+        raise ValueError("recall margin must be >= 0")
+
     fallback_limits = [
         target_faph,
         max(target_faph * 2.0, target_faph + 0.5),
@@ -172,13 +185,27 @@ def _select_best_candidate(
                 return index
         return len(fallback_limits)
 
+    # Stay in the strictest false-accept tier that has a viable candidate. Within
+    # that tier, keep candidates close to the best recall, then spend the allowed
+    # recall margin on the lowest measured false-accept rate.
+    best_tier = min(tier(candidate) for candidate in candidates)
+    tier_candidates = [
+        candidate for candidate in candidates if tier(candidate) == best_tier
+    ]
+    best_recall = max(candidate["recall"] for candidate in tier_candidates)
+    recall_floor = best_recall - recall_margin
+    viable_candidates = [
+        candidate
+        for candidate in tier_candidates
+        if candidate["recall"] >= recall_floor - 1e-12
+    ]
+
     best = min(
-        candidates,
+        viable_candidates,
         key=lambda candidate: (
-            tier(candidate),
-            -candidate["recall"],
             candidate["false_accepts_per_hour"],
-            abs(candidate["sliding_window_size"] - 5),
+            -candidate["recall"],
+            abs(candidate["sliding_window_size"] - PREFERRED_WINDOW_SIZE),
             -candidate["probability_cutoff"],
         ),
     )
@@ -195,7 +222,7 @@ def _load_config(config_path: Path) -> dict:
 
 
 def _load_eval_sets(
-    handler: FeatureHandler,
+    handler: Any,
     config: dict,
 ) -> tuple[str, str, list[np.ndarray], list[np.ndarray]]:
     for positive_mode, ambient_mode in (
@@ -228,7 +255,7 @@ def _load_eval_sets(
 
 
 def _predict_tracks(
-    model: Model,
+    model: Any,
     tracks: Sequence[np.ndarray],
     label: str,
 ) -> list[np.ndarray]:
@@ -244,8 +271,13 @@ def _predict_tracks(
 
 
 def main() -> int:
+    from microwakeword.data import FeatureHandler
+    from microwakeword.inference import Model
+
     args = parse_args()
     window_sizes = _parse_window_sizes(args.window_sizes)
+    if args.recall_margin < 0 or args.recall_margin > 1:
+        raise ValueError("recall-margin must be between 0 and 1")
     if args.cutoff_step <= 0:
         raise ValueError("cutoff-step must be > 0")
     if args.cutoff_max < args.cutoff_min:
@@ -276,6 +308,10 @@ def main() -> int:
         f"→ Evaluating window sizes {window_sizes} with target <= "
         f"{args.target_faph:.2f} false accepts/hour"
     )
+    print(
+        f"→ Favoring lower false accepts within "
+        f"{args.recall_margin:.2%} of the best recall"
+    )
 
     config = _load_config(config_path)
     config["flags"] = config.get("flags", {})
@@ -338,7 +374,11 @@ def main() -> int:
             candidates.append(candidate)
             window_candidates.append(candidate)
 
-        best_window, _ = _select_best_candidate(window_candidates, args.target_faph)
+        best_window, _ = _select_best_candidate(
+            window_candidates,
+            args.target_faph,
+            args.recall_margin,
+        )
         best_by_window.append(best_window)
         print(
             "   window={window}: cutoff={cutoff:.2f}; recall={recall:.2%}; "
@@ -350,7 +390,11 @@ def main() -> int:
             )
         )
 
-    best, selected_limit = _select_best_candidate(candidates, args.target_faph)
+    best, selected_limit = _select_best_candidate(
+        candidates,
+        args.target_faph,
+        args.recall_margin,
+    )
     if best["false_accepts_per_hour"] > args.target_faph + 1e-9:
         print(
             "⚠️  No candidate met the target false accepts/hour budget; "
@@ -390,6 +434,8 @@ def main() -> int:
             "cutoff_min": round(float(cutoffs[0]), 4),
             "cutoff_max": round(float(cutoffs[-1]), 4),
             "cutoff_step": float(args.cutoff_step),
+            "recall_margin": float(args.recall_margin),
+            "preferred_window_size": PREFERRED_WINDOW_SIZE,
         },
         "per_window_best": best_by_window,
         "generated_at": datetime.now(timezone.utc).isoformat(),

cli/setup_blackwell_venv

@@ -0,0 +1,112 @@
+#!/bin/bash
+set -euo pipefail
+
+PROGDIR="$(dirname "$(realpath "$0")")"
+ROOTDIR="$(dirname "${PROGDIR}")"
+
+KNOWN_ARGS=( data-dir force python )
+source "${PROGDIR}/shell.functions"
+
+if [ ${#UNKNOWN_ARGS[@]} -gt 0 ] ; then
+    echo "Unknown argument(s): ${UNKNOWN_ARGS[*]}" >&2
+    HELP=true
+fi
+
+if [ "${HELP}" == "true" ] ; then
+    cat <<EOF >&2
+Usage: setup_blackwell_venv [ --data-dir=/data ] [ --force ] [ --python=python3.13 ]
+
+Creates /data/.venv-blackwell for RTX 50 / Blackwell TensorFlow training.
+Sample generation and augmentation continue to use /data/.venv.
+
+Environment overrides:
+MWW_BLACKWELL_TF_WHEEL_URL: TensorFlow Blackwell wheel URL.
+
+EOF
+    exit 1
+fi
+
+[ -n "${DATA_DIR}" ] && DATA_DIR="$(realpath "${DATA_DIR}")"
+[ -d "${DATA_DIR}" ] || {
+    echo "Data directory '${DATA_DIR}' doesn't exist." >&2
+    exit 1
+}
+
+PYTHON="${PYTHON:-python3.13}"
+VENV="${DATA_DIR}/.venv-blackwell"
+MARKER="${VENV}/.mww-blackwell-venv"
+TF_WHEEL_URL="${MWW_BLACKWELL_TF_WHEEL_URL:-https://github.com/chivitiH/tensorflow-blackwell-python313/releases/download/v2.22.0-selfbuilt/tensorflow-2.22.0.dev0+selfbuilt-cp313-cp313-linux_x86_64.whl}"
+
+if ! command -v "${PYTHON}" >/dev/null 2>&1 ; then
+    echo "Python 3.13 is required for the Blackwell TensorFlow wheel. Missing: ${PYTHON}" >&2
+    exit 1
+fi
+
+if [ "${FORCE:-false}" != "true" ] && [ -x "${VENV}/bin/python" ] && [ -f "${MARKER}" ] ; then
+    echo "   Blackwell TensorFlow venv found (skipping setup_blackwell_venv)"
+    exit 0
+fi
+
+echo "===== Setting up Blackwell TensorFlow environment ${VENV} ====="
+rm -rf "${VENV}" || :
+"${PYTHON}" -m venv --upgrade-deps "${VENV}"
+source "${VENV}/bin/activate"
+
+export PIP_PROGRESS_BAR=off
+export PIP_NO_COLOR=1
+export PIP_QUIET=0
+
+pip_install() {
+    if $VERBOSE ; then
+        pip install "$@" || return 1
+    else
+        { pip install "$@" || return 1 ; } | stdbuf -i0 -o0 tr -d '[:print:]' | stdbuf -i0 -o0 tr '\n' '.'
+    fi
+    echo
+}
+
+echo "   ===== Installing Blackwell TensorFlow wheel ====="
+pip_install --upgrade pip setuptools wheel
+pip_install "${TF_WHEEL_URL}"
+
+echo "   ===== Installing microWakeWord training dependencies ====="
+pip_install \
+    audiomentations \
+    audio_metadata \
+    datasets \
+    mmap_ninja \
+    pymicro-features \
+    pyyaml \
+    webrtcvad-wheels \
+    ai-edge-litert \
+    numpy-minmax \
+    numpy-rms \
+    absl-py \
+    "numpy==2.3.5"
+
+echo "   ===== Checking microwakeword ====="
+MWW="${DATA_DIR}/tools/microWakeWord"
+if [ ! -d "${MWW}" ] || [ -n "$(git -C "${MWW}" status --porcelain 2>/dev/null || true)" ] ; then
+    rm -rf "${MWW}" || :
+    mkdir -p "${DATA_DIR}/tools"
+    echo "   Cloning micro-wake-word to ${DATA_DIR}/tools"
+    git clone https://github.com/TaterTotterson/micro-wake-word "${MWW}" &>/dev/null
+fi
+echo "   Installing microwakeword into Blackwell venv"
+pip_install --no-deps -e "${MWW}"
+
+echo "   ===== Testing Blackwell TensorFlow environment ====="
+"${VENV}/bin/python" - <<'PY'
+import tensorflow as tf
+from ai_edge_litert.interpreter import Interpreter
+from microwakeword.data import FeatureHandler
+from microwakeword.inference import Model
+
+print("TensorFlow:", tf.__version__)
+print("CUDA build:", tf.test.is_built_with_cuda())
+print("GPU:", tf.config.list_physical_devices("GPU"))
+print("microWakeWord Blackwell imports available")
+PY
+
+touch "${MARKER}"
+echo "Blackwell TensorFlow environment ready: ${VENV}"

cli/setup_mit_audio

@@ -25,9 +25,9 @@ fi
 mkdir -p "${DATA_DIR}/training_datasets/downloads" || :
 cd "${DATA_DIR}/training_datasets"
 
-AUDIO_URL="https://mcdermottlab.mit.edu/Reverb/IRMAudio/Audio.zip"
-AUDIO_ZIPFILE="MIT_RIR_Audio.zip"
-AUDIO_ZIP="./downloads/${AUDIO_ZIPFILE}"
+HF_RIR_REPO_ID="TaterTotterson/MIT_environmental_impulse_responses"
+HF_RIR_API_URL="https://huggingface.co/api/datasets/${HF_RIR_REPO_ID}"
+HF_RIR_SOURCE_KEY="hf_mit_environmental_impulse_responses"
 AUDIO_DIR="./mit_rirs"
 mkdir -p "${AUDIO_DIR}" || :
 AUDIO16K_DIR="./mit_rirs_16k"
@@ -35,10 +35,92 @@ mkdir -p "${AUDIO16K_DIR}" || :
 AUDIO_FILECOUNT="./downloads/mit_rir_filecount"
 AUDIO_IN_GLOB="*.wav"
 
-declare -A filecounts=( [${AUDIO_ZIPFILE}]=0 )
+declare -A filecounts=( [${HF_RIR_SOURCE_KEY}]=0 )
 get_filecounts filecounts "${AUDIO_FILECOUNT}"
 
-echo "===== Checking MIT_RIR ====="
+echo "===== Checking MIT environmental RIRs ====="
+
+download_hf_mit_rirs() {
+    source ${DATA_DIR}/.venv/bin/activate
+    python - "${HF_RIR_REPO_ID}" "${HF_RIR_API_URL}" "${AUDIO_DIR}" <<-'EOF'
+import json
+import sys
+import time
+import urllib.parse
+import urllib.request
+from pathlib import Path
+
+repo_id = sys.argv[1]
+api_url = sys.argv[2]
+audio_dir = Path(sys.argv[3])
+audio_dir.mkdir(parents=True, exist_ok=True)
+
+request = urllib.request.Request(api_url, headers={"User-Agent": "WakeWordTrainer/1.0"})
+with urllib.request.urlopen(request, timeout=30) as response:
+    metadata = json.loads(response.read().decode("utf-8"))
+
+files = sorted(
+    sibling.get("rfilename", "")
+    for sibling in metadata.get("siblings", [])
+    if str(sibling.get("rfilename", "")).startswith("16khz/")
+    and str(sibling.get("rfilename", "")).lower().endswith(".wav")
+)
+if not files:
+    raise SystemExit("Hugging Face MIT RIR dataset did not list any 16khz WAV files")
+
+print(f"   Found {len(files)} MIT environmental RIR files on Hugging Face mirror", flush=True)
+downloaded = 0
+skipped = 0
+
+def download_file(url: str, target: Path, rel: str):
+    tmp = target.with_suffix(target.suffix + ".incomplete")
+    for attempt in range(1, 4):
+        try:
+            if tmp.exists():
+                tmp.unlink()
+            with urllib.request.urlopen(url, timeout=30) as response:
+                with tmp.open("wb") as out:
+                    while True:
+                        chunk = response.read(1024 * 64)
+                        if not chunk:
+                            break
+                        out.write(chunk)
+            if not tmp.exists() or tmp.stat().st_size == 0:
+                raise RuntimeError("empty download")
+            tmp.replace(target)
+            return
+        except Exception as exc:
+            if tmp.exists():
+                tmp.unlink()
+            if attempt == 3:
+                raise RuntimeError(f"download failed for {rel}: {exc}") from exc
+            print(f"   Retry {attempt}/2 for {rel}: {exc}", flush=True)
+            time.sleep(2 * attempt)
+
+total = len(files)
+for idx, rel in enumerate(files, start=1):
+    target = audio_dir / rel
+    if target.exists() and target.stat().st_size > 0:
+        skipped += 1
+        if idx == 1 or idx % 25 == 0 or idx == total:
+            print(f"   MIT RIR download progress: {idx}/{total} files ({downloaded} downloaded, {skipped} reused)", flush=True)
+        continue
+    target.parent.mkdir(parents=True, exist_ok=True)
+    encoded = urllib.parse.quote(rel, safe="/")
+    url = f"https://huggingface.co/datasets/{repo_id}/resolve/main/{encoded}"
+    if idx == 1 or idx % 25 == 0 or idx == total:
+        print(f"   Downloading MIT RIR {idx}/{total}: {rel}", flush=True)
+    download_file(url, target, rel)
+    if not target.exists() or target.stat().st_size == 0:
+        raise SystemExit(f"download failed for {rel}")
+    downloaded += 1
+    if idx == 1 or idx % 25 == 0 or idx == total:
+        print(f"   MIT RIR download progress: {idx}/{total} files ({downloaded} downloaded, {skipped} reused)", flush=True)
+
+print(f"   Hugging Face MIT environmental RIR download complete ({downloaded} downloaded, {skipped} reused)", flush=True)
+print(f"   MIT environmental RIR files available: {len(files)}", flush=True)
+EOF
+}
 
 converter() {
     source ${DATA_DIR}/.venv/bin/activate
@@ -58,9 +140,9 @@ rir_out = Path(sys.argv[2])
 
 waves = list(rir_in.rglob("*.wav"))
 try:
-    print("   MIT RIR normalizing to 16k…")
+    print("   MIT environmental RIR normalizing to 16k…")
     # Normalize to 16k mono
-    for p in tqdm(waves, desc="   MIT_RIR (resample 16k mono)"):
+    for p in tqdm(waves, desc="   MIT environmental RIR (resample 16k mono)"):
         outfile = Path(rir_out / p.name)
         if outfile.exists():
             continue
@@ -70,14 +152,14 @@ try:
         if sr != 16000:
             a, _ = librosa.load(p, sr=16000, mono=True)
         write_wav(outfile, a, 16000)
-    print("   MIT RIR normalization complete")
+    print("   MIT environmental RIR normalization complete")
 except Exception as e2:
-    print(f"   MIT RIR fallback failed: {e2}")
+    print(f"   MIT environmental RIR preparation failed: {e2}")
     raise
 EOF
 }
 
-expected_filecount=${filecounts[${AUDIO_ZIPFILE}]}
+expected_filecount=${filecounts[${HF_RIR_SOURCE_KEY}]}
 actual_filecount=$(find "${AUDIO16K_DIR}" -name '*.wav' 2>/dev/null | wc -l) || :
 write_filecount=false
 
@@ -85,24 +167,16 @@ if [ "${actual_filecount}" -ne 0 ] && [ "${actual_filecount}" -eq "${expected_fi
     echo "   Existing ${AUDIO16K_DIR} valid"
 else
     actual_filecount=$(find "${AUDIO_DIR}" -name "${AUDIO_IN_GLOB}" 2>/dev/null | wc -l) || :
-    if [ "${actual_filecount}" -eq 0 ] || [ "${actual_filecount}" -ne "${expected_filecount}" ] ; then
-        if [ ! -f "${AUDIO_ZIP}" ]  ; then
-            echo "   Downloading ${AUDIO_ZIPFILE}"
-            curl -sfL "${AUDIO_URL}" -o "${AUDIO_ZIP}"
-        fi
-
+    if [ "${actual_filecount}" -eq 0 ] || [ "${expected_filecount}" -eq 0 ] || [ "${actual_filecount}" -ne "${expected_filecount}" ] ; then
         rm -rf "${AUDIO_DIR}" || :
-        echo "   Unzipping ${AUDIO_ZIPFILE}"
-        unzip -u -q -d "${AUDIO_DIR}" "${AUDIO_ZIP}"
-    fi
-    if "${CLEANUP_ARCHIVES}" && [ -f "${AUDIO_ZIP}" ] ; then
-        echo "   Cleaning up ${AUDIO_ZIPFILE}"
-        rm -rf "${AUDIO_ZIP}"
+        mkdir -p "${AUDIO_DIR}" || :
+        echo "   Downloading MIT environmental impulse responses from Hugging Face mirror"
+        download_hf_mit_rirs
     fi
 
     converter
     actual_filecount=$(find "${AUDIO16K_DIR}" -name "*.wav" 2>/dev/null | wc -l) || :
-    filecounts[${AUDIO_ZIPFILE}]="${actual_filecount}"
+    filecounts[${HF_RIR_SOURCE_KEY}]="${actual_filecount}"
     write_filecount=true
 fi
 
@@ -110,15 +184,10 @@ if ${write_filecount} ; then
     write_filecounts filecounts "${AUDIO_FILECOUNT}"
 fi
 
-if "${CLEANUP_ARCHIVES}" && [ -f "${AUDIO_ZIP}" ] ; then
-    echo "   Cleaning up ${AUDIO_ZIPFILE}"
-    rm -rf "${AUDIO_ZIP}"
-fi
-
 if "${CLEANUP_INTERMEDIATE_FILES}" && [ -d "${AUDIO_DIR}" ]; then
     echo "   Cleaning up ${AUDIO_DIR}"
     rm -rf "${AUDIO_DIR}"
 fi
 
-echo "   MIT_RIR complete"
+echo "   MIT environmental RIRs complete"
 exit 0

cli/wake_word_sample_augmenter

@@ -18,6 +18,8 @@ parser.add_argument("--output-dir", type=str, help="Wake word output dir. Defaul
 # Personal inputs/outputs (NEW)
 parser.add_argument("--personal-dir", type=str, help="Personal WAV dir. Default: <data-dir>/personal_samples", required=False)
 parser.add_argument("--personal-output-dir", type=str, help="Personal features output dir. Default: <data-dir>/work/personal_augmented_features", required=False)
+parser.add_argument("--negative-dir", type=str, help="Reviewed negative WAV dir. Default: <data-dir>/negative_samples", required=False)
+parser.add_argument("--negative-output-dir", type=str, help="Reviewed negative features output dir. Default: <data-dir>/work/reviewed_negative_features", required=False)
 
 # Dataset dirs
 parser.add_argument("--mit-rirs-16k-dir", type=str, help="MIT RIR input directory. Default: <data-dir>/training_datasets/mit_rirs_16k", required=False)
@@ -57,6 +59,17 @@ if not args.personal_output_dir:
 else:
     args.personal_output_dir = os.path.realpath(args.personal_output_dir)
 
+# Reviewed negative defaults
+if not args.negative_dir:
+    args.negative_dir = os.path.join(args.data_dir, "negative_samples")
+else:
+    args.negative_dir = os.path.realpath(args.negative_dir)
+
+if not args.negative_output_dir:
+    args.negative_output_dir = os.path.join(work_dir, "reviewed_negative_features")
+else:
+    args.negative_output_dir = os.path.realpath(args.negative_output_dir)
+
 # Dataset defaults
 if not args.mit_rirs_16k_dir:
     args.mit_rirs_16k_dir = os.path.join(args.data_dir, "training_datasets", "mit_rirs_16k")
@@ -205,7 +218,7 @@ def bind_wav_generator(clips_obj: Clips, wav_dir: str):
 
     clips_obj.audio_generator = types.MethodType(audio_generator_from_wavs, clips_obj)
 
-def generate_feature_set(input_wav_dir: str, out_root_dir: str, label: str):
+def generate_feature_set(input_wav_dir: str, out_root_dir: str, label: str, *, remove_silence: bool = True):
     files = glob.glob(os.path.join(input_wav_dir, "*.wav"))
     if not files:
         print(f"ℹ️  No WAVs found for {label} in: {input_wav_dir} (skipping)")
@@ -218,7 +231,7 @@ def generate_feature_set(input_wav_dir: str, out_root_dir: str, label: str):
         input_directory=input_wav_dir,
         file_pattern="*.wav",
         max_clip_duration_s=5,
-        remove_silence=True,
+        remove_silence=remove_silence,
         random_split_seed=10,
         split_count=0.1,
     )
@@ -263,9 +276,12 @@ def generate_feature_set(input_wav_dir: str, out_root_dir: str, label: str):
 # Wake word generated/TTS features (existing behavior)
 generate_feature_set(args.input_dir, args.output_dir, "generated")
 
-# Personal features (NEW)
+# Personal features
 generate_feature_set(args.personal_dir, args.personal_output_dir, "personal")
 
+# Reviewed false-positive / hard-negative features
+generate_feature_set(args.negative_dir, args.negative_output_dir, "reviewed negatives", remove_silence=False)
+
 END_TIME = datetime.now(timezone.utc).replace(microsecond=0)
 et = END_TIME - START_TIME
 print(f"\n{'=' * 80}")

cli/wake_word_sample_trainer

@@ -84,6 +84,21 @@ if [ "${IS_BLACKWELL}" = "true" ]; then
   echo "ℹ️  Using GPU compatibility retries; CPU fallback is ${ALLOW_CPU_FALLBACK} (override with MWW_ALLOW_CPU_FALLBACK=true|false)."
 fi
 
+BLACKWELL_TF_MODE="${MWW_BLACKWELL_TF:-auto}"
+BLACKWELL_TF_REQUIRED="false"
+BLACKWELL_TF_ACTIVE="false"
+case "${BLACKWELL_TF_MODE,,}" in
+  1|true|yes|on|required)
+    BLACKWELL_TF_REQUIRED="true"
+    ;;
+  0|false|no|off|disabled)
+    BLACKWELL_TF_MODE="disabled"
+    ;;
+  *)
+    BLACKWELL_TF_MODE="auto"
+    ;;
+esac
+
 # Enable driver-side PTX JIT fallback when ptxas/nvlink are unavailable.
 if [ -z "${XLA_FLAGS:-}" ]; then
   export XLA_FLAGS="--xla_gpu_unsafe_fallback_to_driver_on_ptxas_not_found"
@@ -111,6 +126,16 @@ else
   echo "ℹ️  No personal features found at ${PERSONAL_FEATURES_DIR}/training (continuing without personal weighting)"
 fi
 
+# Reviewed false-positive features are optional hard negatives.
+REVIEWED_NEGATIVE_FEATURES_DIR="${WORK_DIR}/reviewed_negative_features"
+HAS_REVIEWED_NEGATIVE="false"
+if [ -d "${REVIEWED_NEGATIVE_FEATURES_DIR}/training" ] ; then
+  HAS_REVIEWED_NEGATIVE="true"
+  echo "✅ Found reviewed negative features: ${REVIEWED_NEGATIVE_FEATURES_DIR}/training (will weight as hard negatives)"
+else
+  echo "ℹ️  No reviewed negative features found at ${REVIEWED_NEGATIVE_FEATURES_DIR}/training (continuing with stock negatives)"
+fi
+
 cd "${WORK_DIR}"
 
 echo "===== Starting ${TRAINING_STEPS} training steps ====="
@@ -133,6 +158,7 @@ features:
   truth: true
   type: mmap
 __PERSONAL_FEATURE_MARKER__
+__REVIEWED_NEGATIVE_FEATURE_MARKER__
 - features_dir: __NEG_SPEECH__
   penalty_weight: 1.0
   sampling_weight: 12.0
@@ -208,9 +234,51 @@ else
   sed -i -e "/__PERSONAL_FEATURE_MARKER__/d" "${YAML_PATH}"
 fi
 
+# Insert/remove reviewed hard-negative block
+if [ "${HAS_REVIEWED_NEGATIVE}" = "true" ]; then
+  reviewed_negative_block="$(cat <<EOF
+- features_dir: ${REVIEWED_NEGATIVE_FEATURES_DIR}
+  penalty_weight: 1.25
+  sampling_weight: 8.0
+  truncation_strategy: random
+  truth: false
+  type: mmap
+EOF
+)"
+  perl -0777 -i -pe "s#__REVIEWED_NEGATIVE_FEATURE_MARKER__#${reviewed_negative_block}#g" "${YAML_PATH}"
+else
+  sed -i -e "/__REVIEWED_NEGATIVE_FEATURE_MARKER__/d" "${YAML_PATH}"
+fi
+
 echo "   Wrote training_parameters.yaml"
 rm -rf "${WORK_DIR}/trained_models/wakeword"
 
+if [ "${IS_BLACKWELL}" = "true" ] && [ "${BLACKWELL_TF_MODE}" != "disabled" ]; then
+  BLACKWELL_SETUP="${PROGDIR}/setup_blackwell_venv"
+  BLACKWELL_PYTHON="${DATA_DIR}/.venv-blackwell/bin/python"
+
+  if [ -x "${BLACKWELL_SETUP}" ] && command -v python3.13 >/dev/null 2>&1; then
+    echo "↪️  Preparing Blackwell-native TensorFlow training environment."
+    if "${BLACKWELL_SETUP}" --data-dir="${DATA_DIR}"; then
+      PYTHON_BIN="${BLACKWELL_PYTHON}"
+      BLACKWELL_TF_ACTIVE="true"
+      echo "✅ Blackwell TensorFlow training enabled: ${PYTHON_BIN}"
+    else
+      if [ "${BLACKWELL_TF_REQUIRED}" = "true" ]; then
+        echo "❌ Blackwell TensorFlow setup failed and MWW_BLACKWELL_TF is required." >&2
+        exit 1
+      fi
+      echo "⚠️  Blackwell TensorFlow setup failed; continuing with compatibility retries."
+    fi
+  else
+    if [ "${BLACKWELL_TF_REQUIRED}" = "true" ]; then
+      echo "❌ Blackwell TensorFlow was required, but python3.13/setup_blackwell_venv is unavailable." >&2
+      exit 1
+    fi
+    echo "ℹ️  Blackwell TensorFlow image support not available; continuing with compatibility retries."
+  fi
+fi
+
 wake_word_filename="$(
   echo "${WAKE_WORD}" \
     | tr '[:upper:]' '[:lower:]' \
@@ -234,11 +302,11 @@ TRAIN_ARGS=(
   --test_tflite_streaming_quantized 1
   --use_weights best_weights
   mixednet
-  --pointwise_filters "64,64,64,64"
+  --pointwise_filters "128,128,128,128"
   --repeat_in_block "1,1,1,1"
   --mixconv_kernel_sizes "[5], [7,11], [9,15], [23]"
   --residual_connection "0,0,0,0"
-  --first_conv_filters 32
+  --first_conv_filters 64
   --first_conv_kernel_size 5
   --stride 2
 )
@@ -318,6 +386,7 @@ fi
 TRAINING_DONE="false"
 
 echo "🏋️ Starting model training and TFLite export (this is the longest stage)…"
+echo "🧠 Model quality: high_accuracy_plus"
 if run_attempt "Attempt 1/3: GPU training (default runtime profile)" ; then
   echo "✅ Training complete (GPU path)."
   TRAINING_DONE="true"
@@ -398,7 +467,12 @@ echo "🎯 Calibrating detector settings for on-device use…"
 if "${PYTHON_BIN:-python}" "${PROGDIR}/calibrate_detector.py" \
     --training-config "${WORK_DIR}/trained_models/wakeword/training_config.yaml" \
     --model "${source_path}" \
-    --output "${calibration_path}"; then
+    --output "${calibration_path}" \
+    --target-faph "${MWW_CALIBRATION_TARGET_FAPH:-0.25}" \
+    --recall-margin "${MWW_CALIBRATION_RECALL_MARGIN:-0.005}" \
+    --window-sizes "${MWW_CALIBRATION_WINDOW_SIZES:-5,6,7}" \
+    --cutoff-min "${MWW_CALIBRATION_CUTOFF_MIN:-0.95}" \
+    --cutoff-max "${MWW_CALIBRATION_CUTOFF_MAX:-1.00}"; then
     echo "✅ Detector calibration complete."
 else
     echo "⚠️  Detector calibration failed; packaging with default detector settings."
@@ -428,7 +502,9 @@ json_path = Path(os.environ["JSON_PATH"])
 calibration_path = Path(os.environ.get("CALIBRATION_PATH", ""))
 language = (os.environ.get("LANGUAGE", "en") or "en").strip().lower()
 probability_cutoff = 0.97
-sliding_window_size = 5
+sliding_window_size = 6
+strict_min_close_miss_threshold = 0.68
+calibration = {}
 
 if calibration_path.exists():
     try:
@@ -442,21 +518,63 @@ if calibration_path.exists():
     except Exception as exc:
         print(f"⚠️ Failed to read detector calibration ({exc}); using defaults.")
 
+probability_cutoff = round(probability_cutoff, 3)
+sliding_window_size = max(1, min(10, int(sliding_window_size)))
+selected_metrics = calibration.get("selected_metrics") if isinstance(calibration.get("selected_metrics"), dict) else {}
+evaluation = calibration.get("evaluation") if isinstance(calibration.get("evaluation"), dict) else {}
+close_miss_threshold = max(
+    0.01,
+    min(0.99, round(max(strict_min_close_miss_threshold, probability_cutoff - 0.17), 3)),
+)
+
 meta = {
     "type": "micro",
     "wake_word": os.environ["WAKE_WORD_TITLE"],
+    "label": os.environ["WAKE_WORD_TITLE"].replace("_", " ").title(),
     "author": "Tater Totterson",
     "website": "https://github.com/TaterTotterson/microWakeWord-Trainer-Nvidia-Docker.git",
     "model": os.environ["TFLITE_FILENAME"],
     "trained_languages": [language],
     "version": 2,
+    "model_format": "tflite_stream_state_internal_quant",
+    "quantization": "int8",
+    "sample_rate": 16000,
     "micro": {
-        "probability_cutoff": round(probability_cutoff, 2),
+        "probability_cutoff": probability_cutoff,
         "sliding_window_size": sliding_window_size,
         "feature_step_size": 10,
         "tensor_arena_size": 30000,
         "minimum_esphome_version": "2024.7.0",
     },
+    "tater_native": {
+        "format_version": 1,
+        "wake_threshold": probability_cutoff,
+        "wake_sliding_window": sliding_window_size,
+        "close_miss_threshold": close_miss_threshold,
+        "frontend": {
+            "name": "tflm_microfrontend",
+            "sample_rate": 16000,
+            "feature_duration_ms": 30,
+            "feature_step_ms": 10,
+            "feature_size": 40,
+            "input_feature_frames": 2,
+            "lower_band_limit": 125.0,
+            "upper_band_limit": 7500.0,
+        },
+        "recommended_for": ["tater-native-satellite", "voice-pe"],
+    },
+    "calibration": {
+        "target_false_accepts_per_hour": calibration.get("target_false_accepts_per_hour"),
+        "selected_false_accepts_per_hour_limit": calibration.get("selected_false_accepts_per_hour_limit"),
+        "recall": selected_metrics.get("recall"),
+        "false_accepts_per_hour": selected_metrics.get("false_accepts_per_hour"),
+        "ambient_hours": selected_metrics.get("ambient_hours"),
+        "positive_dataset": evaluation.get("positive_dataset"),
+        "ambient_dataset": evaluation.get("ambient_dataset"),
+        "positive_tracks": evaluation.get("positive_tracks"),
+        "ambient_tracks": evaluation.get("ambient_tracks"),
+        "generated_at": calibration.get("generated_at"),
+    },
 }
 json_path.write_text(json.dumps(meta, indent=4) + "\n", encoding="utf-8")
 PY

dockerfile

@@ -6,7 +6,7 @@ ENV DEBIAN_FRONTEND=noninteractive
 # System deps
 RUN apt-get update && apt-get install -y --no-install-recommends \
     python3.12 python3.12-venv python3.12-dev python3-pip python-is-python3 \
-    git wget curl unzip ca-certificates nano less \
+    git wget curl unzip patch ninja-build ca-certificates nano less libgomp1 \
  && rm -rf /var/lib/apt/lists/* \
  && mkdir -p /data
 

dockerfile.blackwell

@@ -0,0 +1,54 @@
+# RTX 50 / Blackwell image
+FROM nvidia/cuda:12.8.1-cudnn-devel-ubuntu24.04
+
+ENV DEBIAN_FRONTEND=noninteractive
+ENV CUDA_HOME=/usr/local/cuda
+ENV PATH=/usr/local/cuda/bin:${PATH}
+ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH}
+ENV MWW_BLACKWELL_IMAGE=1
+ENV MWW_BLACKWELL_TF=auto
+ENV MWW_BLACKWELL_TF_WHEEL_URL=https://github.com/chivitiH/tensorflow-blackwell-python313/releases/download/v2.22.0-selfbuilt/tensorflow-2.22.0.dev0+selfbuilt-cp313-cp313-linux_x86_64.whl
+
+# System deps. Python 3.12 remains the main trainer/runtime venv, while
+# Python 3.13 is used only for the Blackwell TensorFlow training step.
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    software-properties-common ca-certificates curl git wget unzip patch \
+    ninja-build nano less libgomp1 \
+ && add-apt-repository -y ppa:deadsnakes/ppa \
+ && apt-get update \
+ && apt-get install -y --no-install-recommends \
+    python3.12 python3.12-venv python3.12-dev \
+    python3.13 python3.13-venv python3.13-dev \
+    python3-pip python-is-python3 \
+ && ldconfig \
+ && rm -rf /var/lib/apt/lists/* \
+ && mkdir -p /data
+
+# Trainer UI port
+EXPOSE 8789
+
+# Script root
+WORKDIR /root/mww-scripts
+
+# Bash environment
+COPY --chown=root:root --chmod=0755 .bashrc /root/
+
+# Root-level entrypoints
+COPY --chown=root:root --chmod=0755 \
+    train_wake_word \
+    run.sh \
+    trainer_server.py \
+    requirements.txt \
+    /root/mww-scripts/
+
+# CLI folder
+COPY --chown=root:root cli/ /root/mww-scripts/cli/
+
+# Make all CLI scripts executable (avoids "Permission denied")
+RUN chmod -R a+x /root/mww-scripts/cli
+
+# Static UI for trainer
+COPY --chown=root:root --chmod=0644 static/index.html /root/mww-scripts/static/index.html
+
+# trainer server
+CMD ["/bin/bash", "-lc", "/root/mww-scripts/run.sh"]

run.sh

@@ -6,7 +6,7 @@ ROOTDIR="$(dirname "$(realpath "$0")")"
 # Training convention
 DATA_DIR="${DATA_DIR:-/data}"
 HOST="${REC_HOST:-0.0.0.0}"
-PORT="${REC_PORT:-8888}"
+PORT="${REC_PORT:-8789}"
 
 # Keep trainer UI deps separate from the training venv
 VENV_DIR="${DATA_DIR}/.recorder-venv"
@@ -25,6 +25,18 @@ echo "-> URL:      http://localhost:${PORT}/"
 
 mkdir -p "${DATA_DIR}"
 
+install_ui_deps() {
+  ${PIP} install \
+    "fastapi==${FASTAPI_VERSION}" \
+    "uvicorn[standard]==${UVICORN_VERSION}" \
+    "python-multipart==${PY_MULTIPART_VERSION}" \
+    "silero-vad>=5.0.0" \
+    "numpy>=1.24.0" \
+    "faster-whisper>=1.0.0" \
+    "nvidia-cublas-cu12" \
+    "nvidia-cudnn-cu12==9.*"
+}
+
 # -----------------------------
 # Trainer UI venv (separate)
 # -----------------------------
@@ -39,21 +51,89 @@ source "${VENV_DIR}/bin/activate"
 if [[ ! -f "${PIN_FILE}" ]]; then
   echo "Installing pinned trainer UI deps"
   ${PIP} install -U pip setuptools wheel
-  ${PIP} install \
-    "fastapi==${FASTAPI_VERSION}" \
-    "uvicorn[standard]==${UVICORN_VERSION}" \
-    "python-multipart==${PY_MULTIPART_VERSION}"
+  install_ui_deps
   touch "${PIN_FILE}"
 else
   echo "Reusing existing trainer UI venv (no upgrades)"
+  if ! "${PY}" - "${FASTAPI_VERSION}" "${UVICORN_VERSION}" "${PY_MULTIPART_VERSION}" <<'PY' >/dev/null 2>&1
+import importlib.metadata as md
+import sys
+
+fastapi_version, uvicorn_version, multipart_version = sys.argv[1:4]
+
+def version_tuple(value):
+    parts = []
+    for token in str(value).replace("-", ".").split("."):
+        if token.isdigit():
+            parts.append(int(token))
+        else:
+            digits = "".join(ch for ch in token if ch.isdigit())
+            if digits:
+                parts.append(int(digits))
+            break
+    return tuple(parts)
+
+exact = {
+    "fastapi": fastapi_version,
+    "uvicorn": uvicorn_version,
+    "python-multipart": multipart_version,
+}
+minimum = {
+    "silero-vad": "5.0.0",
+    "numpy": "1.24.0",
+    "faster-whisper": "1.0.0",
+    "nvidia-cudnn-cu12": "9.0.0",
+}
+present = (
+    "torch",
+    "nvidia-cublas-cu12",
+)
+
+for package, expected in exact.items():
+    if md.version(package) != expected:
+        raise SystemExit(1)
+for package, minimum_version in minimum.items():
+    if version_tuple(md.version(package)) < version_tuple(minimum_version):
+        raise SystemExit(1)
+for package in present:
+    md.version(package)
+PY
+  then
+    echo "UI dependencies missing or stale; installing recorder dependencies"
+    install_ui_deps
+  fi
 fi
 
+# Faster Whisper/CTranslate2 loads these CUDA libraries before Python starts.
+# They live in the persistent UI venv so both Docker image variants can use GPU STT.
+WHISPER_CUDA_LIBRARY_PATH="$("${PY}" - <<'PY'
+import os
+
+try:
+    import nvidia.cublas.lib
+    import nvidia.cudnn.lib
+except ImportError:
+    print("")
+else:
+    print(
+        os.path.dirname(nvidia.cublas.lib.__file__)
+        + ":"
+        + os.path.dirname(nvidia.cudnn.lib.__file__)
+    )
+PY
+)"
+if [[ -n "${WHISPER_CUDA_LIBRARY_PATH}" ]]; then
+  export LD_LIBRARY_PATH="${WHISPER_CUDA_LIBRARY_PATH}${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}"
+fi
 # -----------------------------
 # Trainer server env
 # -----------------------------
 export DATA_DIR="${DATA_DIR}"
 export STATIC_DIR="${ROOTDIR}/static"
 export PERSONAL_DIR="${DATA_DIR}/personal_samples"
+export CAPTURED_DIR="${DATA_DIR}/captured_audio"
+export NEGATIVE_DIR="${DATA_DIR}/negative_samples"
+export TRAINED_WAKE_WORDS_DIR="${DATA_DIR}/trained_wake_words"
 
 # IMPORTANT: leave training venv creation to /api/train inside trainer_server.py
 # but still set TRAIN_CMD so the server knows how to invoke training once ready

tests/test_auto_train.py

@@ -0,0 +1,238 @@
+import io
+import json
+import sys
+import tempfile
+import unittest
+import wave
+from pathlib import Path
+from types import SimpleNamespace
+from unittest.mock import Mock, patch
+
+import trainer_server as trainer
+
+
+def silent_wav_bytes(duration_s: float = 0.25) -> bytes:
+    output = io.BytesIO()
+    with wave.open(output, "wb") as wav_file:
+        wav_file.setnchannels(1)
+        wav_file.setsampwidth(2)
+        wav_file.setframerate(16000)
+        wav_file.writeframes(b"\x00\x00" * int(16000 * duration_s))
+    return output.getvalue()
+
+
+class AutoTrainTests(unittest.TestCase):
+    def setUp(self):
+        self.tempdir = tempfile.TemporaryDirectory()
+        root = Path(self.tempdir.name)
+        self.original_paths = (
+            trainer.CAPTURED_DIR,
+            trainer.NEGATIVE_DIR,
+            trainer.PERSONAL_DIR,
+            trainer.AUTO_TRAIN_CONFIG_FILE,
+            trainer.AUTO_TRAIN_STATE_FILE,
+        )
+        trainer.CAPTURED_DIR = root / "captured_audio"
+        trainer.NEGATIVE_DIR = root / "negative_samples"
+        trainer.PERSONAL_DIR = root / "personal_samples"
+        trainer.AUTO_TRAIN_CONFIG_FILE = root / "auto_train_config.json"
+        trainer.AUTO_TRAIN_STATE_FILE = root / "auto_train_state.json"
+        for directory in (trainer.CAPTURED_DIR, trainer.NEGATIVE_DIR, trainer.PERSONAL_DIR):
+            directory.mkdir(parents=True)
+
+        self.original_config = dict(trainer.AUTO_TRAIN_CONFIG)
+        self.original_state = dict(trainer.AUTO_TRAIN_STATE)
+        trainer.AUTO_TRAIN_CONFIG.clear()
+        trainer.AUTO_TRAIN_CONFIG.update(
+            trainer._normalize_auto_train_config(
+                {
+                    "enabled": True,
+                    "wake_phrase": "hey tater",
+                    "language": "en",
+                    "tater_url": "http://127.0.0.1:8501",
+                    "stt_device": "auto",
+                    "stt_compute_type": "auto",
+                }
+            )
+        )
+        trainer.AUTO_TRAIN_STATE.clear()
+        trainer.AUTO_TRAIN_STATE.update(trainer.AUTO_TRAIN_DEFAULT_STATE)
+
+    def tearDown(self):
+        (
+            trainer.CAPTURED_DIR,
+            trainer.NEGATIVE_DIR,
+            trainer.PERSONAL_DIR,
+            trainer.AUTO_TRAIN_CONFIG_FILE,
+            trainer.AUTO_TRAIN_STATE_FILE,
+        ) = self.original_paths
+        trainer.AUTO_TRAIN_CONFIG.clear()
+        trainer.AUTO_TRAIN_CONFIG.update(self.original_config)
+        trainer.AUTO_TRAIN_STATE.clear()
+        trainer.AUTO_TRAIN_STATE.update(self.original_state)
+        self.tempdir.cleanup()
+
+    def add_capture(self, name: str = "wake.wav", wake_word: str = "hey_tater") -> Path:
+        audio_path = trainer.CAPTURED_DIR / name
+        audio_path.write_bytes(silent_wav_bytes())
+        trainer._write_sidecar_json(
+            audio_path,
+            {
+                "original_name": name,
+                "wake_word": wake_word,
+                "event_type": "wake_detected",
+                "review_status": "pending",
+            },
+        )
+        return audio_path
+
+    def test_phrase_matching_normalizes_case_punctuation_and_underscores(self):
+        self.assertTrue(trainer._transcript_contains_wake_phrase("Okay, HEY TATER!", "hey_tater"))
+        self.assertFalse(trainer._transcript_contains_wake_phrase("Turn on the television", "hey tater"))
+
+    def test_phrase_miss_moves_wake_trigger_to_negative_samples(self):
+        self.add_capture()
+        with patch.object(trainer, "_transcribe_capture_with_faster_whisper", return_value="turn on the kitchen lights"):
+            trainer._auto_review_capture("wake.wav")
+
+        self.assertFalse((trainer.CAPTURED_DIR / "wake.wav").exists())
+        negatives = list(trainer.NEGATIVE_DIR.glob("*.wav"))
+        self.assertEqual(len(negatives), 1)
+        metadata = trainer._load_sidecar_json(negatives[0])
+        self.assertTrue(metadata["auto_negative"])
+        self.assertEqual(metadata["review_status"], "auto_approved_negative")
+        self.assertEqual(metadata["transcript"], "turn on the kitchen lights")
+        self.assertEqual(trainer.AUTO_TRAIN_STATE["pending_negative_count"], 1)
+
+    def test_matching_phrase_stays_in_manual_review_inbox(self):
+        audio_path = self.add_capture()
+        with patch.object(trainer, "_transcribe_capture_with_faster_whisper", return_value="hey tater turn on the lights"):
+            trainer._auto_review_capture("wake.wav")
+
+        self.assertTrue(audio_path.exists())
+        self.assertFalse(list(trainer.NEGATIVE_DIR.glob("*.wav")))
+        metadata = trainer._load_sidecar_json(audio_path)
+        self.assertEqual(metadata["auto_review_status"], "wake_phrase_detected")
+        self.assertEqual(trainer.AUTO_TRAIN_STATE["pending_negative_count"], 0)
+
+    def test_capture_for_another_wake_word_is_not_transcribed(self):
+        audio_path = self.add_capture(wake_word="computer")
+        with patch.object(trainer, "_transcribe_capture_with_faster_whisper") as transcribe:
+            trainer._auto_review_capture("wake.wav")
+
+        transcribe.assert_not_called()
+        self.assertTrue(audio_path.exists())
+        metadata = trainer._load_sidecar_json(audio_path)
+        self.assertEqual(metadata["auto_review_status"], "different_wake_phrase")
+
+    def test_due_schedule_starts_training_after_minimum_negatives(self):
+        trainer.AUTO_TRAIN_CONFIG["schedule_hours"] = 24
+        trainer.AUTO_TRAIN_CONFIG["minimum_new_negatives"] = 3
+        trainer.AUTO_TRAIN_STATE["pending_negative_count"] = 3
+        trainer.AUTO_TRAIN_STATE["next_run_at"] = "2000-01-01T00:00:00+00:00"
+        with patch.object(trainer, "_start_auto_training", return_value={"ok": True, "started": True}) as start:
+            trainer._maybe_run_scheduled_auto_training()
+
+        start.assert_called_once_with()
+        self.assertTrue(trainer.AUTO_TRAIN_STATE["next_run_at"])
+
+    def test_tater_refresh_repushes_settings_with_selector_and_token(self):
+        trainer.AUTO_TRAIN_CONFIG.update(
+            {
+                "notify_satellites": True,
+                "tater_url": "http://127.0.0.1:8501",
+                "tater_selector": "kitchen-sat",
+                "tater_api_token": "secret-token",
+            }
+        )
+
+        class Response:
+            def __enter__(self):
+                return self
+
+            def __exit__(self, *_args):
+                return False
+
+            def read(self):
+                return b'{"push":{"count":2}}'
+
+        with patch.object(trainer, "urlopen", return_value=Response()) as open_url:
+            result = trainer._notify_tater_satellites()
+
+        self.assertTrue(result["ok"])
+        self.assertEqual(result["count"], 2)
+        request = open_url.call_args.args[0]
+        self.assertEqual(request.full_url, "http://127.0.0.1:8501/api/tater/satellite/v1/settings")
+        self.assertEqual(request.get_header("X-tater-token"), "secret-token")
+        self.assertEqual(json.loads(request.data), {"selector": "kitchen-sat", "settings": {}})
+
+    def test_advertised_url_uses_non_loopback_browser_host(self):
+        request = SimpleNamespace(
+            base_url="http://192.168.1.50:8789/",
+            url=SimpleNamespace(hostname="192.168.1.50", scheme="http", port=8789),
+        )
+        self.assertEqual(trainer._advertised_base_url(request), "http://192.168.1.50:8789")
+
+    def test_advertised_url_replaces_localhost_with_discovered_lan_host(self):
+        request = SimpleNamespace(
+            base_url="http://127.0.0.1:8789/",
+            url=SimpleNamespace(hostname="127.0.0.1", scheme="http", port=8789),
+        )
+        with patch.object(trainer, "_discover_lan_ipv4", return_value="192.168.1.60"):
+            self.assertEqual(trainer._advertised_base_url(request), "http://192.168.1.60:8789")
+
+    def test_configured_public_url_takes_precedence(self):
+        trainer.AUTO_TRAIN_CONFIG["advertised_base_url"] = "http://trainer.local:8789"
+        request = SimpleNamespace(
+            base_url="http://127.0.0.1:8789/",
+            url=SimpleNamespace(hostname="127.0.0.1", scheme="http", port=8789),
+        )
+        self.assertEqual(trainer._advertised_base_url(request), "http://trainer.local:8789")
+
+    def test_faster_whisper_auto_runtime_prefers_cuda_and_float16(self):
+        fake_ctranslate2 = SimpleNamespace(get_cuda_device_count=lambda: 1)
+        with patch.dict(sys.modules, {"ctranslate2": fake_ctranslate2}):
+            self.assertEqual(
+                trainer._resolve_faster_whisper_runtime("auto", "auto"),
+                ("cuda", "float16"),
+            )
+
+    def test_faster_whisper_auto_runtime_falls_back_to_cpu_int8(self):
+        fake_ctranslate2 = SimpleNamespace(get_cuda_device_count=lambda: 0)
+        with patch.dict(sys.modules, {"ctranslate2": fake_ctranslate2}):
+            self.assertEqual(
+                trainer._resolve_faster_whisper_runtime("auto", "auto"),
+                ("cpu", "int8"),
+            )
+
+    def test_faster_whisper_transcription_joins_segments_and_records_runtime(self):
+        fake_model = SimpleNamespace()
+        fake_model.transcribe = Mock(
+            return_value=(
+                iter([SimpleNamespace(text=" turn on "), SimpleNamespace(text="the lights ")]),
+                SimpleNamespace(),
+            )
+        )
+        with (
+            patch.object(trainer, "_resolve_faster_whisper_runtime", return_value=("cuda", "float16")),
+            patch.object(trainer, "_load_faster_whisper_model", return_value=fake_model),
+        ):
+            transcript = trainer._transcribe_capture_with_faster_whisper(
+                Path("wake.wav"),
+                model="small.en",
+                language="en",
+            )
+
+        self.assertEqual(transcript, "turn on the lights")
+        fake_model.transcribe.assert_called_once_with(
+            "wake.wav",
+            language="en",
+            beam_size=1,
+            condition_on_previous_text=False,
+        )
+        self.assertEqual(trainer.AUTO_TRAIN_STATE["last_stt_device"], "cuda")
+        self.assertEqual(trainer.AUTO_TRAIN_STATE["last_stt_compute_type"], "float16")
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_calibrate_detector.py

@@ -0,0 +1,88 @@
+import importlib.util
+import unittest
+from pathlib import Path
+
+
+SCRIPT_PATH = (
+    Path(__file__).resolve().parents[1]
+    / "cli"
+    / "calibrate_detector.py"
+)
+SPEC = importlib.util.spec_from_file_location("calibrate_detector", SCRIPT_PATH)
+calibrate_detector = importlib.util.module_from_spec(SPEC)
+assert SPEC.loader is not None
+SPEC.loader.exec_module(calibrate_detector)
+
+
+def candidate(cutoff, window, recall, false_accepts_per_hour):
+    return {
+        "probability_cutoff": cutoff,
+        "sliding_window_size": window,
+        "recall": recall,
+        "false_accepts_per_hour": false_accepts_per_hour,
+    }
+
+
+class CalibrationSelectionTests(unittest.TestCase):
+    def test_defaults_are_conservative(self):
+        self.assertEqual(calibrate_detector.DEFAULT_WINDOW_SIZES, [5, 6, 7])
+        self.assertEqual(calibrate_detector.DEFAULT_CUTOFF_MIN, 0.95)
+        self.assertEqual(calibrate_detector.DEFAULT_RECALL_MARGIN, 0.005)
+
+    def test_prefers_zero_false_accepts_within_recall_margin(self):
+        candidates = [
+            candidate(0.95, 5, 0.99894, 0.103408),
+            candidate(0.95, 6, 0.99744, 0.0),
+            candidate(0.95, 7, 0.99554, 0.0),
+        ]
+
+        best, selected_limit = calibrate_detector._select_best_candidate(
+            candidates,
+            target_faph=0.25,
+            recall_margin=0.005,
+        )
+
+        self.assertEqual(best["sliding_window_size"], 6)
+        self.assertEqual(best["false_accepts_per_hour"], 0.0)
+        self.assertEqual(selected_limit, 0.25)
+
+    def test_does_not_trade_away_recall_beyond_margin(self):
+        candidates = [
+            candidate(0.95, 5, 0.99, 0.1),
+            candidate(0.99, 6, 0.90, 0.0),
+        ]
+
+        best, _ = calibrate_detector._select_best_candidate(
+            candidates,
+            target_faph=0.25,
+            recall_margin=0.005,
+        )
+
+        self.assertEqual(best["sliding_window_size"], 5)
+
+    def test_uses_strictest_available_false_accept_tier(self):
+        candidates = [
+            candidate(0.95, 5, 0.99, 0.6),
+            candidate(0.99, 6, 0.99, 1.5),
+        ]
+
+        best, selected_limit = calibrate_detector._select_best_candidate(
+            candidates,
+            target_faph=0.25,
+            recall_margin=0.005,
+        )
+
+        self.assertEqual(best["false_accepts_per_hour"], 0.6)
+        self.assertEqual(selected_limit, 0.75)
+
+    def test_rejects_negative_recall_margin(self):
+        with self.assertRaises(ValueError):
+            calibrate_detector._select_best_candidate(
+                [candidate(0.95, 6, 0.99, 0.0)],
+                target_faph=0.25,
+                recall_margin=-0.001,
+            )
+
+
+if __name__ == "__main__":
+    unittest.main()

train_wake_word

@@ -150,6 +150,7 @@ if ${CLEANUP_WORK_DIR} ; then
       "${DATA_DIR}/work/wake_word_samples" \
       "${DATA_DIR}/work/wake_word_samples_augmented" \
       "${DATA_DIR}/work/personal_augmented_features" \
+      "${DATA_DIR}/work/reviewed_negative_features" \
       "${DATA_DIR}/work/last_wake_word" || :
 fi