| --- |
| title: "WakeForge: GCP & Piper TTS Wake Word Dataset Creator" |
| emoji: 🔨 |
| colorFrom: indigo |
| colorTo: purple |
| sdk: gradio |
| sdk_version: 6.19.0 |
| app_file: app.py |
| pinned: false |
| license: cc-by-4.0 |
| --- |
| |
| # 🔨 WakeForge |
| ### GCP & Piper TTS Wake Word Dataset Creator |
|
|
| A Hugging Face Space that generates a synthetic **keyword-spotting / wake-word** |
| dataset ready for **Hugging Face Datasets** and **Edge Impulse**. |
|
|
| It uses **Google Cloud Text-to-Speech** when you provide an API key, and |
| **automatically falls back to free, local Piper TTS** when you don't — so it |
| always works, with or without a paid service. |
|
|
| ## Features |
|
|
| - 🔀 **Automatic backend selection** — Google Cloud TTS (API key) with a free |
| Piper TTS fallback. |
| - 🧱 **Three keyword-spotting classes** — `hey_android`, `unknown`, |
| `background_noise` (fully configurable phrases and labels). |
| - 🎛️ **Local augmentation** — gain, time shift, additive noise, echo, plus |
| synthetic background-noise generation. |
| - 📦 **Edge Impulse-ready** — `label.<id>.wav` filenames and `training` / |
| `testing` folders. |
| - ⬆️ **One-click publishing** — push to a Hugging Face dataset repo and/or |
| **upload directly to your Edge Impulse project** with your own API key. |
|
|
| ## Using the Space |
|
|
| 1. Set the phrases, labels and dataset size. |
| 2. *(Optional)* Paste a **Google Cloud TTS API key**. Leave blank to use free |
| Piper TTS. |
| 3. *(Optional)* Tick **Push to Hugging Face** and provide a repo id + write token. |
| 4. *(Optional)* Tick **Upload to Edge Impulse** and paste your project API key |
| (Edge Impulse → your project → **Dashboard → Keys**). |
| 5. Click **Generate dataset** and download the resulting zip. |
|
|
| ### Space secrets (optional) |
|
|
| Instead of typing keys into the UI, set them as Space secrets: |
|
|
| | Secret | Purpose | |
| |---|---| |
| | `GCP_TTS_API_KEY` | Google Cloud TTS API key | |
| | `HF_TOKEN` | Hugging Face write token for pushing datasets | |
| | `EDGE_IMPULSE_API_KEY` | Edge Impulse project API key | |
|
|
| ## Command line |
|
|
| The same pipeline runs locally without Gradio: |
|
|
| ```bash |
| pip install -r requirements.txt |
| |
| # Free Piper TTS, prepare an HF folder locally |
| python generate.py --out output --hf-out hf_dataset |
| |
| # Google Cloud TTS + push to HF + upload to Edge Impulse |
| python generate.py \ |
| --gcp-api-key "$GCP_TTS_API_KEY" \ |
| --push-hf-repo "username/hey-android" --hf-token "$HF_TOKEN" \ |
| --edge-impulse-api-key "$EDGE_IMPULSE_API_KEY" |
| ``` |
|
|
| ## Output layout |
|
|
| ```text |
| output/ |
| edge_impulse_upload/ |
| training/ hey_android.<id>.wav ... |
| testing/ hey_android.<id>.wav ... |
| by_label/ |
| hey_android/ ... |
| metadata.csv |
| selected_voices.csv |
| dataset_summary.json |
| |
| hf_dataset/ |
| audio/train/ ... |
| audio/test/ ... |
| README.md # dataset card |
| hf_metadata.csv |
| metadata.csv |
| ``` |
|
|
| ## Getting the keys |
|
|
| - **Google Cloud TTS API key** — [Google Cloud Console](https://console.cloud.google.com/): |
| enable *Cloud Text-to-Speech API*, then **APIs & Services → Credentials → |
| Create credentials → API key**. |
| - **Edge Impulse API key** — your project → **Dashboard → Keys**. |
| - **Hugging Face token** — [huggingface.co/settings/tokens](https://huggingface.co/settings/tokens) |
| (needs *write* access). |
|
|
| ## Notes & limitations |
|
|
| Synthetic TTS audio is great for **bootstrapping** a wake-word model but is not |
| a production benchmark. Before deploying, add real recordings from the target |
| device and expected acoustic environments. |
|
|
| ## License |
|
|
| CC BY 4.0. Verify that your use of generated synthetic speech complies with the |
| terms of the voice models (Google Cloud TTS / Piper voices) you use. |
|
|
