Adding a Provider to AudioBard

AudioBard is designed with an anti-fragile, pluggable architecture. All external engines (LLM providers, TTS synthesis engines, and book parsers) are hidden behind abstract base classes.

---

1. Adding a new LLM Provider

To add a new LLM provider (e.g. Anthropic, Mistral, LocalAI):

1. Subclass LLMClient in src/audiobard/llm/<provider>_client.py:

   from audiobard.llm.base import LLMClient
   from typing import Any
   
   class CustomLLMClient(LLMClient):
       def __init__(self, model: str = "default-model", **kwargs: Any) -> None:
           super().__init__(model=model, **kwargs)
   
       async def _raw_call(self, prompt: str, schema: dict[str, Any]) -> str:
           # Call provider API with JSON constrained schema
           ...
           return raw_json_string
   

2. Register the provider in src/audiobard/pipeline.py:

3. Update create_llm_client(config, persistence) factory to recognize the new provider identifier.

4. Update AudioBardConfig.llm_provider type annotations in src/audiobard/config.py.

5. Declare Optional Dependencies:

6. Add extra dependencies to [project.optional-dependencies] in pyproject.toml.

7. Add Unit Tests:

8. Add unit tests with mock HTTP responses (using respx or unittest.mock) in tests/test_llm_<provider>.py.

---

2. Adding a new TTS Provider

To add a new TTS provider (e.g. Coqui TTS, Kokoro, ElevenLabs):

1. Subclass TTSProvider in src/audiobard/tts/<provider>_provider.py:

   from audiobard.tts.base import TTSProvider
   from audiobard.models import Voice, Emotion
   
   class CustomTTSProvider(TTSProvider):
       async def _synthesize_raw(
           self,
           text: str,
           voice: Voice,
           emotion: Emotion,
           rate: float,
           pitch: float,
       ) -> bytes:
           # Synthesize audio and return raw MP3 bytes
           ...
           return mp3_bytes
   
       async def list_voices(self, locale: str) -> list[Voice]:
           # Return available Voice models for locale
           ...
           return [Voice(id="voice-id", locale=locale, ...)]
   

2. Inherent Caching & Rate Limiting:

3. TTSProvider automatically handles in-memory LRU caching (500 MB limit) and disk caching in ~/.cache/audiobard/tts/.

4. Concurrency is automatically throttled by the provider semaphore (config.tts_semaphore).

5. Register the Provider:

6. Update create_tts_provider(config) factory in src/audiobard/pipeline.py.

7. Add the option to AudioBardConfig.tts_provider.

8. Add Unit Tests:

9. Test synthesis arguments and voice listing in tests/test_tts_<provider>.py.