feat: update TTS to bark-small and other updates in associated project documentation, added links required for submission

README.md

@@ -9,35 +9,53 @@ python_version: '3.10'
 app_file: app.py
 pinned: false
 tags:
-  - track:Thousand-Token-Wood
-  - sponsor:OpenBMB
-  - sponsor:OpenAI
-  - bonus:Tiny-Titan
+  - track:wood
+  - sponsor:openbmb
+  - sponsor:openai
+  - achievement:offgrid
+  - bonus:tiny-titan
 ---
 
 # Samantic Audio
 
 Semantic-to-audio communication system that features a local robotic AI. 
 
-The application uses an LLM to generate responses, converts the response to speech using Kokoro TTS, and processes the output through a custom synthesizer to create a unique "droid-style" voice.
+The application uses an LLM to generate responses, converts the response to speech using bark-small TTS, and processes the output through a custom synthesizer to create a unique "droid-style" voice.
+
+## Links
+
+| Resource | Link |
+| :--- | :--- |
+| **Demo Video** | [YouTube](https://youtu.be/ID0IG2BIBRo) |
+| **Social Post** | [X Post](https://x.com/GovindLokam/status/2066625722891026578?s=20) |
+| **Social Post** | [LinkedIn Post](https://www.linkedin.com/posts/govind-lokam-335382230_ai-llm-generativeai-share-7472390184009895936-6YOz/?utm_source=share&utm_medium=member_desktop&rcm=ACoAADm4O1kBjwPZCkmLC-YSFR8At-SNQhkj4XY) |
+| **GitHub Repository** | [GovIndLok/Voinal](https://github.com/GovIndLok/Voinal) |
+
+## Hackathon Badges Claimed
+
+| Category | Badge Name (Tag) | Description / Justification |
+| :--- | :--- | :--- |
+| **Track** | Thousand Token Wood (`track:wood`) | Whimsical, delightful, AI-native app |
+| **Sponsor** | MiniCPM Build (`sponsor:openbmb`) | Used MiniCPM5-1B for LLM |
+| **Sponsor** | Codex (`sponsor:openai`) | Codex-attributed commits in the connected GitHub repo |
+| **Achievement** | Off the Grid (`achievement:offgrid`) | Both models run in-Space on ZeroGPU; no external AI API is called. And can also be run locally as shown even in demo |
+| **Bonus** | Tiny-titan (`bonus:tiny-titan`) | Models must be ≤ 4B parameters. (1B (minicpm5) + ~240M (bark-small) <= 4B) |
 
 ## Features
-- **Local LLM Integration**: Uses Ollama with `qwen3.5:4b` to generate responses locally.
-- **Text-to-Speech**: Uses the Kokoro TTS model for voice generation.
+- **Local LLM Integration**: Uses `MiniCPM5-1B` to generate responses locally.
+- **Text-to-Speech**: Uses the `bark-small` TTS model for voice generation.
 - **Custom Audio Synthesis**: The `synth.py` pipeline transforms audio into a droid-style output.
 - **Web Interface**: A Gradio web UI to easily chat with the robot.
 
 ## Requirements
 - Python >= 3.12
-- Local Ollama server installed and running.
-- `espeak-ng` installed and available on the PATH.
 
 ## Setup & Installation
 Activate your virtual environment and install the required packages:
 
 ```bash
 source .venv/bin/activate
-pip install -r pyproject.toml # or install dependencies manually: gradio ollama numpy scipy torch kokoro soundfile
+pip install -r pyproject.toml # or install dependencies manually: gradio numpy scipy torch soundfile
 ```
 
 ## Running the Application
@@ -52,5 +70,5 @@ The application will be served at `http://127.0.0.1:7860`.
 ## Project Structure
 - `app.py`: Gradio entrypoint serving the UI and handling the full pipeline.
 - `synth.py`: Audio-processing pipeline to transform standard WAVs into droid-style output.
-- `tts_model.py`: Kokoro TTS integration.
+- `tts_model.py`: `bark-small` TTS integration.
 - `pyproject.toml`: Project metadata and dependencies.

synth.py

@@ -119,7 +119,7 @@ def droid_synth_array(fs, voice_data, droid_type="sml"):
 
     for i in range(total_sample):
         if sample_until_tick <= 0:
-            if droid_type == "r2":
+            if droid_type == "sml":
                 interval = random.choice([1.0, 1.25, 1.5, 2.0, 0.75])
                 target = np.clip(current_freq * interval, *p["freq_range"])
             else: