Fastyr AI Pipeline

A flexible and extensible Python library for building AI-powered conversational pipelines. Fastyr provides a clean interface between Speech-to-Text (STT), Language Model (LLM), and Text-to-Speech (TTS) services.

Features

• Seamless integration of STT, LLM, and TTS services
• Easy provider switching with consistent interfaces
• Async/await support for optimal performance
• Built-in error handling and retries
• Flexible storage backend support

Installation

pip install fastyr-ai-pipeline

Quick Start

import os
from fastyr.services.providers.pipeline_service import PipelineService
from fastyr.services.providers.deepgram_provider import DeepgramProvider
from fastyr.services.providers.openai_provider import OpenAIProvider
from fastyr.services.providers.elevenlabs_provider import ElevenLabsProvider
from fastyr.services.providers.local_storage_provider import LocalStorageProvider

# Initialize providers
storage = LocalStorageProvider(base_path="storage/audio")
pipeline = PipelineService(
    stt_provider=DeepgramProvider(api_key=os.getenv('DEEPGRAM_API_KEY')),
    llm_provider=OpenAIProvider(api_key=os.getenv('OPENAI_API_KEY')),
    tts_provider=ElevenLabsProvider(
        api_key=os.getenv('ELEVENLABS_API_KEY'),
        voice_id="your-voice-id"
    ),
    storage_provider=storage
)

# Process audio through the pipeline
result = await pipeline.process(request)
print(f"Processed audio available at: {result.audio_url}")

Provider Interfaces

The library defines three core interfaces:

• STTProvider: transcribe(audio_data: bytes, options: Dict) -> str
• LLMProvider: generate_response(prompt: str, options: Dict) -> str
• TTSProvider: synthesize(text: str, options: Dict) -> bytes

Implemented Providers

• Deepgram (STT): High-accuracy speech recognition with multi-language support
• OpenAI (LLM): GPT-3.5/4 integration with customizable prompts
• ElevenLabs (TTS): High-quality voice synthesis with multiple voice options

Configuration

Set environment variables for your API keys:

DEEPGRAM_API_KEY=your_deepgram_key
OPENAI_API_KEY=your_openai_key
ELEVENLABS_API_KEY=your_elevenlabs_key

Custom Providers

Implement your own provider by extending the relevant interface:

from fastyr.services.interfaces.stt_provider import STTProvider
from typing import Dict, Any

class CustomSTTProvider(STTProvider):
    async def transcribe(self, audio_data: bytes, options: Dict[str, Any] = None) -> str:
        # Your implementation
        pass

Requirements

• Python 3.7+
• See setup.py for dependencies

License

MIT License - see LICENSE file for details

Author

William Jefferson Mensah - GitHub