[FEATURE] Add Native Audio Support to ContentBlock Type

[westonbrown]

Problem Statement

The Strands SDK's ContentBlock type currently supports image, video, and document content but lacks audio support. As multimodal AI models increasingly support audio input/output, this gap forces developers to use untyped workarounds, breaking type safety and SDK consistency.

The recently merged LlamaCpp provider (PR #585) demonstrates this limitation - it must cast to Dict[str, Any] to handle audio for models like Qwen2.5-Omni, losing the type safety that makes Strands reliable.

Proposed Solution

Proposed Solution

Extend the SDK's media types to include audio, following the established pattern for other media types:

# src/strands/types/media.py
AudioFormat = Literal["wav", "mp3", "flac"]

class AudioSource(TypedDict):
    """Contains the content of audio data."""
    bytes: bytes

class AudioContent(TypedDict):
    """Audio to include in a message."""
    format: AudioFormat
    source: AudioSource

# src/strands/types/content.py
class ContentBlock(TypedDict, total=False):
    # ... existing fields ...
    audio: AudioContent  # Add alongside image, video, document

Use Case

This enhancement would benefit:

• Model Providers: Bedrock (Nova Sonic), LlamaCpp (Qwen2.5-Omni)
• Applications: Voice assistants, transcription services, audio analysis, real-time conversation

Alternatives Solutions

No response

Additional Context

The LlamaCpp implementation currently handles audio as:

# Current workaround in llamacpp.py
if "audio" in content:
    audio_content = cast(Dict[str, Any], content)  # Loss of type safety
    audio_data = base64.b64encode(audio_content["audio"]["source"]["bytes"])

With native support, all model providers could handle audio consistently and safely within the SDK's type system.