AutoAgents Speech

Speech (TTS/STT) provider abstractions for the AutoAgents framework.

This crate provides trait-based abstraction layers for speech providers, allowing different backends to be used interchangeably within the AutoAgents ecosystem.

Features

TTS (Text-to-Speech)

• Speech Generation: Generate audio from text
• Voice Management: Use predefined voices
• Streaming Support: Optional streaming for real-time audio generation
• Model Management: Support for multiple models and languages

STT (Speech-to-Text)

• Transcription: Convert audio to text
• Streaming Support: Real-time audio transcription
• Timestamp Support: Token-level timestamps for transcriptions
• Multilingual: Support for multiple languages with auto-detection

Architecture

The crate follows a trait-based design with provider implementations in the providers module:

TTS Traits

• TTSProvider: Marker trait combining all TTS capabilities
• TTSSpeechProvider: Speech generation capabilities
• TTSModelsProvider: Model and language support

STT Traits

• STTProvider: Marker trait combining all STT capabilities
• STTSpeechProvider: Transcription capabilities
• STTModelsProvider: Model and language support

Providers

Enable providers using feature flags:

• pocket-tts: Pocket-TTS model support (TTS)
• parakeet: Parakeet (NVIDIA) model support (STT)
• vad: Silero VAD support (speech segmentation)