[BUG] [BidiAgent/BidiOpenAIRealtimeModel] openai transcription failed

[fvisticot]

Checks

• I have updated to the lastest minor and patch version of Strands
• I have checked the documentation and this is not expected behavior
• I have searched ./issues and there are no duplicates of my issue

Strands Version

1.20.0

Python Version

3.12

Operating System

macOS

Installation Method

pip

Steps to Reproduce

import asyncio
import logging
import os

from strands.experimental.bidi import BidiAgent
from strands.experimental.bidi.io import BidiAudioIO, BidiTextIO
from strands.experimental.bidi.models.openai_realtime import BidiOpenAIRealtimeModel
from strands.experimental.bidi.tools import stop_conversation

from strands_tools import calculator

# Load environment variables from .env file
from dotenv import load_dotenv
load_dotenv()

# Configure logging
logging.basicConfig(
    level=logging.INFO,
    format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
    handlers=[logging.StreamHandler()]
)
logger = logging.getLogger(__name__)

# Set strands to WARNING to reduce noise
logging.getLogger("strands").setLevel(logging.INFO)


async def main() -> None:
    logger.info("🎤 Starting OpenAI Voice Agent...")

    # Get OpenAI API key
    api_key = os.environ.get('OPENAI_API_KEY')
    if not api_key:
        logger.error("OPENAI_API_KEY environment variable not set")
        raise ValueError("OPENAI_API_KEY environment variable is required")
    
    logger.info("✅ API key loaded from environment")

    logger.info("🔧 Creating BidiOpenAIRealtimeModel...")
    model = BidiOpenAIRealtimeModel(
        model_id="gpt-realtime",
        provider_config={
            "audio": {
                "input_rate": 24000,
                "output_rate": 24000,
                "channels": 1,
                "format": "pcm",
                "voice": "coral",
            },
        },
        client_config={
            "api_key": api_key,
        },
    )
    logger.info("✅ Model created successfully")

    # stop_conversation tool allows user to verbally stop agent execution.
    logger.info("🤖 Creating BidiAgent with tools: calculator, stop_conversation")
    agent = BidiAgent(model=model, tools=[calculator, stop_conversation])
    logger.info("✅ Agent created successfully")

    logger.info("🎧 Setting up audio I/O...")
    audio_io = BidiAudioIO()
    text_io = BidiTextIO()
    logger.info("✅ Audio I/O ready")

    logger.info("🚀 Starting voice conversation...")
    logger.info("💡 Say 'stop conversation' to end the session")
    logger.info("⌨️  Press Ctrl+C for emergency stop")
    
    try:
        await agent.run(inputs=[audio_io.input()], outputs=[audio_io.output(), text_io.output()])
    except asyncio.CancelledError:
        logger.info("🛑 Session cancelled by user")
    except Exception as e:
        logger.error(f"❌ Error during conversation: {e}", exc_info=True)
        raise
    finally:
        logger.info("👋 Voice session ended")


if __name__ == "__main__":
    try:
        asyncio.run(main())
    except KeyboardInterrupt:
        logger.info("🛑 Interrupted by user (Ctrl+C)")
    except Exception as e:
        logger.error(f"❌ Fatal error: {e}", exc_info=True)

Expected Behavior

Audio and transcript works

Actual Behavior

interrupted
2025-12-20 13:28:39,725 - strands.experimental.bidi.models.openai_realtime - WARNING - error=<Input transcription failed for item 'item_CoqMYmtK90wuTOQVid1yI'.> | openai transcription failed
interrupted
2025-12-20 13:28:44,726 - strands.experimental.bidi.models.openai_realtime - WARNING - error=<Input transcription failed for item 'item_CoqMcUAxpSAmrVn4KALN6'.> | openai transcription failed

Additional Context

API KEY is correct

Possible Solution

No response

Related Issues

No response

[bluemeaford]

Thanks for the clear repro + logs. The warning suggests the realtime session is receiving audio items, but the transcription step is failing upstream.

This may be a configuration or capability issue rather than a Strands bug, so a bit more detail would help narrow it down.

A couple things that would help clarify whether this is Strands vs config/model access:
1. Can you enable DEBUG for strands.experimental.bidi.models.openai_realtime (or strands) and paste the full error payload/trace around the “openai transcription failed” warning?
2. Which model_id is expected here? Is gpt-realtime from the Strands docs/examples, and does switching to the example model_id change behavior?
3. Can you confirm the audio format Strands is sending (pcm16 vs pcm, sample rate) matches your input device?